StatAILab/DataSciEval
2
1[2 {3 "data_type": "tabular data",4 "domain": "Data Analysis",5 "task_type": "Exploratory Data Analysis, Statistical Testing & Inference",6 "language": "Python",7 "question": "Are there any meaningful numeric correlations within the Bloomington accidents dataset that could inform predictive modeling or accident investigation strategies?",8 "reasoning": "After loading the dataset, only numeric columns should be identified and extracted since correlation analysis operates on numeric data. The correlation matrix should be computed to measure the strength and direction of linear relationships between all numeric features. The correlations should be visualized using a heatmap with color coding and annotated values to make patterns easily identifiable. The analysis should identify which numeric features show strong positive or negative correlations with each other, as these relationships could indicate interdependencies or hidden patterns in accident data. Strong correlations could suggest that certain features carry redundant information, while weak or zero correlations across the numeric space suggest that the predictive power lies mainly in the categorical features. This analysis informs feature selection and helps understand the underlying structure of the data.",9 "answer": "The heatmap shows only a single numeric variable, 'cfs_number', with a self-correlation of 1.0. No other numeric columns appear in the correlation matrix, so there are no pairwise numeric correlations visible that could inform modeling. This implies either the dataset has very few numeric fields or numeric features were not included in the correlation plot. To proceed, extract or engineer additional numeric features (or encode relevant categorical/temporal fields such as month, weekday, responders, area) before expecting informative numeric correlations for predictive modeling.",10 "confidence": 2.0,11 "notebook": "bloomington-traffic-accident-analysis-and-predi.ipynb",12 "id": 35,13 "figure": "<image_id:3>",14 "dataset_size_mb": 2.402237892150879,15 "dataset": "melissamonfared/bloomington-accidents"16 },17 {18 "data_type": "tabular data",19 "domain": "Data Analysis",20 "task_type": "Data Ingestion & Integration, Reporting & Interpretation",21 "language": "Python",22 "question": "I have the CSV file 'bestsellers with categories.csv' containing Amazon's Top 50 bestselling books from 2009 to 2019. Starting from the raw file, report the dataset dimensions, the inferred data types of each column, compute the percentage of missing values per column, and visualize the missingness matrix.",23 "reasoning": "Begin by loading the CSV into a tabular structure to confirm the columns and data are read correctly. Determine the dataset shape to understand how many records and columns are present. Inspect the inferred data types for each column to separate categorical and numeric variables. For data completeness, compute the proportion of missing values for every column; this indicates whether further cleaning is required. Finally, render a missingness matrix to visually verify the absence or presence of gaps across rows and columns.",24 "answer": "Dimensions: 550 records and 7 columns. Data types: Name (object), Author (object), User Rating (float64), Reviews (int64), Price (int64), Year (int64), Genre (object). Missing values per column: Name - 0.0%, Author - 0.0%, User Rating - 0.0%, Reviews - 0.0%, Price - 0.0%, Year - 0.0%, Genre - 0.0%. Missingness visualization confirms no missing values. <image_id:0>",25 "confidence": 4.0,26 "notebook": "amazon-s-books-eda-plotly-hypothesis-test.ipynb",27 "id": 45,28 "figure": "<image_id:0>",29 "dataset_size_mb": 0.048790931701660004,30 "dataset": "sootersaalu/amazon-top-50-bestselling-books-2009-2019"31 },32 {33 "data_type": "tabular data",34 "domain": "Data Analysis",35 "task_type": "Data Preparation & Wrangling",36 "language": "Python",37 "question": "Given the same raw Amazon bestsellers CSV, standardize title casing for book names, harmonize author spellings using fuzzy matching (e.g., unify 'George R. R. Martin' with 'George R.R. Martin' and 'J. K. Rowling' with 'J.K. Rowling'), drop the 'Year' column, remove duplicate books keeping the latest entry by title, and report the final counts of distinct books and authors.",38 "reasoning": "Read the raw data, then normalize string formatting (title case and trimming spaces) to reduce string-level duplicates in Name and Author. Use fuzzy string matching to identify and unify semantically identical author spellings (for example the variants of George R.R. Martin and J.K. Rowling). Remove the 'Year' column to avoid multiple rows for the same title across different years. Identify duplicate rows and then remove duplicates by book title, retaining the last occurrence to reflect the most recent pricing where titles repeat across years. Finally, report the post-cleaning dataset size and the unique counts of books and authors.",39 "answer": "Author unique values decreased from 248 to 246 after harmonizing spellings. After dropping the 'Year' column and an initial duplicate removal, the dataset had 361 entries across 6 columns. Because some titles (e.g., 'The Help') were duplicated with differing prices across years, dropping duplicates by 'Name' while keeping the last entry produced the final dataset with 350 rows and 6 columns. Final counts: 350 distinct books written by 246 authors.",40 "confidence": 3.0,41 "notebook": "amazon-s-books-eda-plotly-hypothesis-test.ipynb",42 "id": 46,43 "figure": null,44 "dataset_size_mb": 0.048790931701660004,45 "dataset": "sootersaalu/amazon-top-50-bestselling-books-2009-2019"46 },47 {48 "data_type": "tabular data",49 "domain": "Data Analysis",50 "task_type": "Data Preparation & Wrangling, Exploratory Data Analysis",51 "language": "Python",52 "question": "From the cleaned Amazon bestsellers data, identify which authors have written the most bestsellers and list their counts.",53 "reasoning": "Aggregate the data by author, counting the number of unique book titles attributed to each author to measure their bestseller frequency. Rank authors by this count in descending order to reveal the top contributors. Report the top authors and their corresponding counts.",54 "answer": "Top authors by number of bestsellers: Jeff Kinney — 12 books; Rick Riordan — 10 books; J.K. Rowling — 8 books; Stephenie Meyer — 7 books; Dav Pilkey — 6 books; Bill O'Reilly — 6 books; John Grisham — 5 books; E L James — 5 books; Suzanne Collins — 5 books; Charlaine Harris — 4 books.",55 "confidence": 3.0,56 "notebook": "amazon-s-books-eda-plotly-hypothesis-test.ipynb",57 "id": 47,58 "figure": null,59 "dataset_size_mb": 0.048790931701660004,60 "dataset": "sootersaalu/amazon-top-50-bestselling-books-2009-2019"61 },62 {63 "data_type": "tabular data",64 "domain": "Statistical Testing & Experimentation",65 "task_type": "Statistical Testing & Inference",66 "language": "Python",67 "question": "Using the cleaned dataset of Amazon bestsellers (after deduplication), compute descriptive statistics for the numeric columns (User Rating, Reviews, Price) and the Pearson correlation matrix among these variables.",68 "reasoning": "Summarize the distribution of each numeric variable by calculating count, mean, standard deviation, minimum, quartiles, and maximum to understand central tendency and variability. Then compute the pairwise Pearson correlations between User Rating, Reviews, and Price to quantify linear relationships. Present the exact computed values.",69 "answer": "Descriptive statistics (count, mean, std, min, 25%, 50%, 75%, max): User Rating — 350.0, 4.608857, 0.226993, 3.3, 4.50, 4.6, 4.80, 4.9; Reviews — 350.0, 9804.605714, 10885.017686, 37.0, 3435.25, 6328.0, 11510.25, 87841.0; Price — 350.0, 12.925714, 10.003161, 0.0, 7.25, 11.0, 16.00, 105.0. Pearson correlation matrix: User Rating with Reviews = -0.055478, User Rating with Price = -0.028228, Reviews with Price = -0.045705.",70 "confidence": 4.0,71 "notebook": "amazon-s-books-eda-plotly-hypothesis-test.ipynb",72 "id": 48,73 "figure": null,74 "dataset_size_mb": 0.048790931701660004,75 "dataset": "sootersaalu/amazon-top-50-bestselling-books-2009-2019"76 },77 {78 "data_type": "tabular data",79 "domain": "Statistical Testing & Experimentation",80 "task_type": "Model Evaluation & Selection, Statistical Testing & Inference",81 "language": "Python",82 "question": "Starting from the cleaned Amazon bestsellers data, test whether user ratings differ between the two genres (Non Fiction vs Fiction). First check normality of User Rating, then apply an appropriate two-sample test and report the test statistics, p-values, and median ratings by genre.",83 "reasoning": "First, verify whether User Rating follows a normal distribution using the Shapiro–Wilk test. Given non-normality, split the data into two groups by Genre and apply a nonparametric two-sample test (Mann–Whitney U) to compare their distributions without assuming normality. Evaluate the p-value against a 0.05 significance level to accept or reject the null hypothesis of no difference. Finally, compare medians to interpret directionality of any difference.",84 "answer": "Normality check (Shapiro–Wilk) on User Rating: Statistic = 0.877, P-Value = 0.00000000000000045166 → reject normality. Mann–Whitney U test between Non Fiction and Fiction: U = 13013.000, P-Value = 0.00931903468696572077 → reject the null; there are significant differences between genres. Median ratings: Non Fiction = 4.6, Fiction = 4.7.",85 "confidence": 4.0,86 "notebook": "amazon-s-books-eda-plotly-hypothesis-test.ipynb",87 "id": 49,88 "figure": null,89 "dataset_size_mb": 0.048790931701660004,90 "dataset": "sootersaalu/amazon-top-50-bestselling-books-2009-2019"91 },92 {93 "data_type": "tabular data, text data",94 "domain": "Natural Language Processing",95 "task_type": "Data Preparation & Wrangling, Feature Engineering & Preparation, Reporting & Interpretation",96 "language": "Python",97 "question": "Using the Animal Crossing user_reviews.csv dataset, split reviews into Negative and Positive categories based on the dataset’s median grade. Clean the review text by lowercasing, removing punctuation, stop words, and the words 'animal' and 'crossing'. Then compute and visualize the top 15 most frequent words for each category.",98 "reasoning": "Load the raw dataset and determine the median of the numeric grade to define a threshold that separates lower-scored reviews (Negative) from higher-scored reviews (Positive). Partition the reviews accordingly. For each subset, clean the text: remove punctuation and URLs, convert to lowercase, and remove common stop words to focus on meaningful tokens; also remove domain-specific terms ('animal', 'crossing') to avoid them dominating the counts. Tokenize the cleaned text and compute word frequencies. Sort the frequencies, select the top 15 words in each category, and visualize them using horizontal bar charts to compare the most common tokens between Negative and Positive reviews.",99 "answer": "The two horizontal bar charts show the top 15 most frequent words for Negative and Positive reviews, respectively. In both charts the word 'game' is the most frequent (around 4,000–4,500 occurrences). Negative reviews rank 'island' and 'one' next (island ≈2,200–2,400; one ≈1,700–1,900) followed by words like 'play', 'switch', 'player', 'nintendo', 'per', 'console', 'first', 'cant', 'expand', 'buy', 'get', 'like'. Positive reviews also include 'island' and 'one' but at lower counts (island ≈1,400–1,600; one ≈1,100–1,300) and show words such as 'new', 'like', 'switch', 'play', 'expand', 'people', 'time', 'per', 'really', 'player', 'dont', 'fun'. There is overlap in many common words (e.g., game, island, one, switch, play, expand, player), but the relative ranks and counts differ: several words (like 'island' and 'one') appear noticeably more often in Negative reviews, while Positive reviews include words such as 'new', 'people', 'time', 'really', and 'fun' that do not appear in the Negative top-15.",100 "confidence": 3.0,101 "notebook": "shifterator-analysis-on-animal-crossing-reviews.ipynb",102 "id": 90,103 "figure": "<image_id:1> <image_id:2>",104 "dataset_size_mb": 2.710921287536621,105 "dataset": "jessemostipak/animal-crossing"106 },107 {108 "data_type": "tabular data, text data",109 "domain": "Natural Language Processing",110 "task_type": "Data Preparation & Wrangling, Exploratory Data Analysis, Reporting & Interpretation",111 "language": "Python",112 "question": "Starting from the Animal Crossing user_reviews.csv dataset, create cleaned word frequency distributions for Negative and Positive review corpora and use Shifterator to construct an entropy shift graph quantifying which words drive the differences between the two sets. Provide the resulting visualization and summarize what the graph encodes.",113 "reasoning": "Load the dataset and split the reviews into Negative and Positive groups using the median grade as the boundary. Clean both corpora by removing punctuation, common stop words, and domain-specific tokens, and convert text to lowercase. Count word frequencies for each group to obtain comparable distributions. Use an entropy-based word shift approach to contrast the distributions and quantify each word’s contribution to the difference. Render the shift graph to visualize which words increase or decrease entropy between Negative (reference) and Positive (comparison) review texts, with colors indicating which corpus a word is associated with.",114 "answer": "An entropy shift graph comparing Negative (reference) and Positive (comparison) reviews was produced. The visualization shows word-level contributions to the difference in usage, with Negative-associated contributions displayed in purple and Positive-associated contributions in yellow. Specific top contributing words are visible on the chart. <image_id:4>",115 "confidence": 3.0,116 "notebook": "shifterator-analysis-on-animal-crossing-reviews.ipynb",117 "id": 92,118 "figure": "<image_id:4>",119 "dataset_size_mb": 2.710921287536621,120 "dataset": "jessemostipak/animal-crossing"121 },122 {123 "data_type": "image data",124 "domain": "Computer Vision",125 "task_type": "Data Ingestion & Integration",126 "language": "Python",127 "question": "Using the split Boot/Sandal/Shoe dataset, create normalized image iterators at 128x128 resolution with batch size 32. Report the number of images detected in each split and the class index mapping used.",128 "reasoning": "Create normalized iterators that read images so that each subfolder corresponds to a class. When the iterators are instantiated, they report the number of images found per split and infer the class-to-index mapping from subfolder names. These outputs let us confirm split sizes and label encoding prior to training.",129 "answer": "Found 12000 images belonging to 3 classes.\n\nFound 1500 images belonging to 3 classes.\n\nFound 1500 images belonging to 3 classes.\n\nClass indices for training generator: {'Boot': 0, 'Sandal': 1, 'Shoe': 2}",130 "confidence": 4.0,131 "notebook": "shoe-vs-sandal-vs-boot-multiclass-acc-0-98.ipynb",132 "id": 128,133 "figure": null,134 "dataset_size_mb": 47.491047859191895,135 "dataset": "hasibalmuzdadid/shoe-vs-sandal-vs-boot-dataset-15k-images"136 },137 {138 "data_type": "image data",139 "domain": "Computer Vision",140 "task_type": "Model Training & Optimization",141 "language": "Python",142 "question": "From the raw image classification setup (Boot/Sandal/Shoe) with 128x128x3 inputs, define a CNN classifier ending with a 3-unit softmax output and report the parameter counts (total, trainable, non-trainable) from the model summary.",143 "reasoning": "Design a sequential convolutional network appropriate for 128x128 color images, including convolution, pooling, normalization, dropout, and dense layers, and finalize with a 3-unit softmax for the three classes. After defining the model, review the model summary to obtain exact totals for all parameters, distinguishing trainable and non-trainable counts.",144 "answer": "Model name: Abdullah_CNN\nTotal params: 1,846,815\nTrainable params: 1,846,559\nNon-trainable params: 256",145 "confidence": 4.0,146 "notebook": "shoe-vs-sandal-vs-boot-multiclass-acc-0-98.ipynb",147 "id": 129,148 "figure": null,149 "dataset_size_mb": 47.491047859191895,150 "dataset": "hasibalmuzdadid/shoe-vs-sandal-vs-boot-dataset-15k-images"151 },152 {153 "data_type": "image data",154 "domain": "Computer Vision",155 "task_type": "Prediction & Forecasting, Reporting & Interpretation",156 "language": "Python",157 "question": "From the raw image classification setup (Boot/Sandal/Shoe) with 128x128x3 inputs. Using CNN, randomly sample 20 images and visualize a 4x5 grid showing each image with its true and predicted label.",158 "reasoning": "Load the test images via a normalized iterator with single-image batches to maintain label alignment. For a sample of images, feed each through the trained model to obtain a class probability vector, convert it to the predicted class via the highest probability, and place both the true and predicted labels as titles of the displayed images arranged in a grid.",159 "answer": "The image displays a 4x5 grid (20 test images) of footwear, each annotated with 'True: <class>' and 'Predicted: <class>'. The visible classes are Shoe, Sandal, and Boot. In this grid all 20 shown images have matching true and predicted labels (no visible misclassifications). The statement 'Found 1500 images' is not supported by the displayed visualization, which only shows 20 images.",160 "confidence": 4.0,161 "notebook": "shoe-vs-sandal-vs-boot-multiclass-acc-0-98.ipynb",162 "id": 131,163 "figure": "<image_id:6>",164 "dataset_size_mb": 47.491047859191895,165 "dataset": "hasibalmuzdadid/shoe-vs-sandal-vs-boot-dataset-15k-images"166 },167 {168 "data_type": "image data",169 "domain": "Computer Vision",170 "task_type": "Data Ingestion & Integration, Data Preparation & Wrangling",171 "language": "Python",172 "question": "I have an American Sign Language (ASL) image dataset stored by class (36 classes). Starting from the raw data, split the data into train, validation, and test sets using an 80/10/10 ratio, load them with an image data generator (200x200 RGB, batch size 32), and report how many images and classes are found in each split.",173 "reasoning": "Begin by reading the raw image that contains one subfolder per class. Perform a stratified split into training, validation, and testing with the target ratios so each split maintains class structure. Use an image data generator to standardize images by rescaling and to load them from the split directories with a fixed target size and batch size. When initializing the directory iterators for train, validation, and test, they will enumerate how many images and how many distinct class folders they contain. Report these counts to verify the splits and class coverage.",174 "answer": "Train: 2012 images belonging to 36 classes. Validation: 251 images belonging to 36 classes. Test: 252 images belonging to 36 classes.",175 "confidence": 4.0,176 "notebook": "hand-sign-multi-class-classification-cnn-97.ipynb",177 "id": 132,178 "figure": null,179 "dataset_size_mb": 61.920475006103516,180 "dataset": "ayuraj/asl-dataset"181 },182 {183 "data_type": "image data",184 "domain": "Computer Vision",185 "task_type": "Model Training & Optimization, Reporting & Interpretation",186 "language": "Python",187 "question": "From the raw ASL images (loaded at 200×200×3), define and summarize a CNN with three convolutional blocks (32/64/128 filters, ReLU, max pooling, dropouts), followed by fully connected layers (512 and 128 units with dropouts) and a 36-way softmax. Report the total and trainable parameter counts from the model summary.",188 "reasoning": "Construct a sequential convolutional architecture that ingests 200×200 RGB images. Build three convolutional blocks with increasing filter counts, each followed by pooling and dropout to reduce spatial size and control overfitting. Flatten the feature maps and add dense layers with ReLU activations and dropouts to learn non-linear combinations of features. Finish with a softmax layer sized to the number of classes. Summarize the model to compute the total parameters and confirm all parameters are trainable.",189 "answer": "Total params: 41,317,828; Trainable params: 41,317,828; Non-trainable params: 0.",190 "confidence": 4.0,191 "notebook": "hand-sign-multi-class-classification-cnn-97.ipynb",192 "id": 134,193 "figure": null,194 "dataset_size_mb": 61.920475006103516,195 "dataset": "ayuraj/asl-dataset"196 },197 {198 "data_type": "image data",199 "domain": "Computer Vision",200 "task_type": "Model Evaluation & Selection, Model Training & Optimization",201 "language": "Python",202 "question": "Train CNN on the raw ASL dataset using an image data generator, with early stopping and learning-rate reduction on plateau, for up to 30 epochs. What is the best validation accuracy achieved during training, and how did the learning rate adjust during training?",203 "reasoning": "Load the train and validation splits with consistent preprocessing and batch sizes. Compile the CNN with a suitable optimizer and categorical loss. Fit the model while monitoring validation metrics for early stopping and learning-rate scheduling. Track validation accuracy across epochs and identify its maximum to assess generalization. Also record when the scheduler reduces the learning rate upon plateauing validation performance.",204 "answer": "Best validation accuracy achieved: 97.61% (val_accuracy = 0.9761 at epoch 14). Learning rate was reduced on plateaus at epoch 7 to 0.0005, epoch 10 to 0.00025, and epoch 13 to 0.000125.",205 "confidence": 4.0,206 "notebook": "hand-sign-multi-class-classification-cnn-97.ipynb",207 "id": 135,208 "figure": null,209 "dataset_size_mb": 61.920475006103516,210 "dataset": "ayuraj/asl-dataset"211 },212 {213 "data_type": "tabular data",214 "domain": "Data Analysis",215 "task_type": "Reporting & Interpretation, Statistical Testing & Inference",216 "language": "Python",217 "question": "Given the academic stress dataset with temporal information, what were the peak stress hours identified through time series analysis, and what statistical evidence supports the correlation between these hours and stress levels?",218 "reasoning": "First, the dataset would be transformed to include time-based features such as hour of day, day of week, and time of day. Next, time series decomposition would be performed to identify patterns and trends in stress levels over time. Rolling statistics would be calculated to smooth the data and identify significant fluctuations. Statistical tests such as t-tests would be conducted to compare stress levels during different time periods. Finally, confidence intervals would be calculated for hourly stress data to determine the most significant stress peaks.",219 "answer": "The plots indicate the highest mean stress values occur in the early morning, with the hourly mean curve peaking around roughly 6–8 AM (the largest point is near hour 8 with a mean ≈4.2–4.5). However, the hourly error bars (mean ± 95% CI) are wide and overlap neighboring hours, so the morning rise is modest and not sharply distinct. The figure does not display a reported Pearson r or p-value for a time-of-day vs stress correlation; the only explicit test shown is a weekend vs weekday t-test (t=0.163, p=0.871), which is non‑significant. Also note the response-rate histogram shows a large spike in responses around hour 23, indicating uneven sampling across hours that could bias hourly means. In summary: visual peaks appear at ~6–8 AM, but the plotted 95% CIs overlap and no clear, reported statistical correlation (r/p) linking those specific hours to stress levels is shown in the images.",220 "confidence": 4.0,221 "notebook": "student-stress-analysis-with-ai-models.ipynb",222 "id": 159,223 "figure": "<image_id:2>",224 "dataset_size_mb": 0.013628005981445,225 "dataset": "poushal02/student-academic-stress-real-world-dataset"226 },227 {228 "data_type": "tabular data",229 "domain": "Clustering",230 "task_type": "Feature Engineering & Preparation, Pattern & Anomaly Detection",231 "language": "Python",232 "question": "Given the academic stress dataset with multiple pressure factors, how many distinct student clusters were identified through advanced clustering analysis, and what were the key characteristics that defined each cluster in terms of stress levels and academic pressure factors?",233 "reasoning": "First, a comprehensive feature set would be constructed to represent student stress profiles, including both original and engineered features. Next, dimensionality reduction techniques would be applied to visualize the data in lower dimensions. Then, multiple clustering algorithms would be applied to identify natural groupings in the data. The optimal number of clusters would be determined through silhouette score analysis and elbow method. Finally, each cluster would be characterized by its average feature values, with particular attention to stress levels and pressure factors, to understand their practical implications for academic stress management.",234 "answer": "The dashboard shows 4 distinct student clusters (labels 0–3). Key characteristics visible in the Cluster Characteristics (normalized radar) and supporting plots are: \n- Cluster 0: a balanced/moderate profile across peer pressure, academic pressure, competition and stress — a mid-stress group. \n- Cluster 1: the highest normalized Stress_Index and Total_Pressure (peaks on those axes) and elevated academic pressure — the most highly stressed cluster. \n- Cluster 2: low normalized values across most pressure factors and stress — the lowest-stress group. \n- Cluster 3: relatively high Peer_Pressure and Academic_Competition but lower Stress_Index/Total_Pressure than Cluster 1 — students under social/competitive pressure without the highest overall stress. \nThe PCA and t-SNE plots show these clusters are mostly separable with some overlap, and the feature-importance plot indicates peer and academic pressure are among the top contributors to the clustering.",235 "confidence": 4.0,236 "notebook": "student-stress-analysis-with-ai-models.ipynb",237 "id": 160,238 "figure": "<image_id:4>",239 "dataset_size_mb": 0.013628005981445,240 "dataset": "poushal02/student-academic-stress-real-world-dataset"241 },242 {243 "data_type": "tabular data",244 "domain": "Business Analytics",245 "task_type": "Prediction & Forecasting, Reporting & Interpretation",246 "language": "Python",247 "question": "Given the academic stress dataset. Analysis the stress risk segmentation, what strategic recommendations would you make for targeting interventions to reduce academic stress among high-risk students, and what ROI can be expected from implementing these recommendations?",248 "reasoning": "First, the student population would be segmented into different risk levels based on stress indices. Next, the characteristics of high-risk students would be identified through analysis of key factors like academic pressure, coping strategies, and environmental factors. Then, intervention strategies would be designed based on these characteristics to address the specific needs of high-risk students. Finally, the cost-effectiveness of these interventions would be evaluated to determine the expected ROI, considering both implementation costs and potential benefits from reduced stress levels.",249 "answer": "Key observations from the dashboard (visuals only):\n- Population in scope: 63.6% of students fall into the top risk bands when combining High (40.0%) and Critical (23.6%).\n- Recommended resource allocation by risk segment (from the “Recommended Resource Allocation” chart): prioritize Critical (≈40% of resources), High (≈35%), Moderate (≈20%), Low (≈5%).\n- Intervention priority (from the Intervention Priority Matrix): Mental Health Services and Stress Management Training sit in the high-impact / high-feasibility quadrant and should be prioritized first. Study environment improvements and Family Counseling have moderate impact/feasibility; Peer Support and Time Management appear lower on impact (even if feasible).\n- Estimated ROI by intervention (from the ROI bar chart): Time Management ≈250%, Stress Management ≈220%, Environment Improvement ≈200%, Peer Support ≈192%, Family Counseling ≈175%.\n- Caveat: the models/metrics panel shows a low model accuracy (~20%), so results should be validated with pilots and monitored closely.\nPractical, dashboard-consistent recommendations:\n1) Immediate priority: roll out Stress Management Training and Mental Health Services to high- and critical-risk students (allocate the bulk of program resources to Critical ≈40% and High ≈35% segments). These interventions are shown as high impact/feasible and Stress Management also has one of the highest ROIs (~220%).\n2) Parallel pilots: implement Study Environment improvements (ROI ≈200%) and Family Counseling (ROI ≈175%) in a subset of schools/ cohorts to validate effectiveness given their moderate priority.\n3) Targeted low-cost programs: deploy Time Management and Peer Support more broadly but as lower-priority complements — Time Management has the highest single-intervention ROI (~250%) but appears lower on impact in the priority matrix, so use it where feasible and low-cost.\n4) Monitoring & validation: because model accuracy is low (~20%), run controlled pilots, track intervention success rate and early-warning effectiveness (dashboard KPI panel), and reallocate resources based on measured outcomes.\nExpected ROI (dashboard figures): individual interventions show high ROIs (Time Management ~250%, Stress Management ~220%, Environment Improvement ~200%, Peer Support ~192%, Family Counseling ~175%). The dashboard does not provide a single combined ROI percentage, an explicit dollar investment, or a break-even timeline for the combined program — those figures in the original answer (138.4% ROI, $70k investment, $166,875 benefit, 8.7 months break-even) are not shown on the provided visuals and cannot be verified from these images.",250 "confidence": 4.0,251 "notebook": "student-stress-analysis-with-ai-models.ipynb",252 "id": 161,253 "figure": "<image_id:5>",254 "dataset_size_mb": 0.013628005981445,255 "dataset": "poushal02/student-academic-stress-real-world-dataset"256 },257 {258 "data_type": "tabular data",259 "domain": "Data Analysis",260 "task_type": "Data Preparation & Wrangling, Exploratory Data Analysis",261 "language": "Python",262 "question": "From the raw IMDB Top 250 data, compute the frequency of each content certificate category and identify the most common certificate.",263 "reasoning": "Load the dataset and focus on the categorical 'certificate' column. Count occurrences of each unique certificate value to obtain a frequency distribution. Sort or directly interpret the counts to find which certificate appears most frequently.",264 "answer": "Certificate counts:\nR: 97\nPG: 37\nPG-13: 35\nNot Rated: 24\nG: 19\nPassed: 16\nApproved: 14\n18+: 1\nNot Available: 1\nTV-PG: 1\nUnrated: 1\nX: 1\n13+: 1\nTV-MA: 1\nGP: 1\nMost common certificate: R (97).",265 "confidence": 4.0,266 "notebook": "imdb-movies-analysis-eda-recommendations.ipynb",267 "id": 222,268 "figure": null,269 "dataset_size_mb": 0.10676097869873001,270 "dataset": "rajugc/imdb-top-250-movies-dataset"271 },272 {273 "data_type": "tabular data",274 "domain": "Data Analysis",275 "task_type": "Exploratory Data Analysis, Feature Engineering & Preparation, Reporting & Interpretation",276 "language": "Python",277 "question": "Using World Cup summary data from 1930 to 2022, can you analyze the relationship between hosting a World Cup tournament and winning it, and determine which continents have hosted tournaments most frequently and whether continental location influences championship outcomes?",278 "reasoning": "First, the World Cup summary dataset must be loaded containing information about host countries and champions for each tournament year. A binary feature should be engineered by comparing the host country with the champion country for each tournament year to create a 'HOST WINNER' indicator variable, marking True when a host nation won the championship and False otherwise. Next, the host countries should be mapped to their respective continents using geographical knowledge or reference data. This continental mapping should categorize host nations into groups such as Europe, South America, North America, Africa, and Asia. Similarly, champion countries should be mapped to their continents to enable continental-level analysis. The frequency distribution of host countries by continent should be calculated to determine which continents have hosted tournaments most frequently. Concurrently, the frequency distribution of champions by continent should be analyzed. The relationship between host wins and tournament outcomes should be examined to understand whether hosting provides a competitive advantage. Visualizations should be created to compare tournament hosting distribution and championship distribution across continents.",279 "answer": "Images show: Hosts per continent — Europe 10 hosts (45.45%), South America 5 (22.73%), North America 3 (13.64%), Asia 3 (13.64%), Africa 1 (4.55%). Champion distribution by country — Brazil 5 titles (22.73%), Italy 4 (18.18%), Germany 4 (18.18%), Argentina 3 (13.64%), Uruguay 2 (9.09%), France 2 (9.09%), England 1 (4.55%), Spain 1 (4.55%). Champions by continent — Europe 12 titles (54.55%) and South America 10 titles (45.45%). The provided images do not show a host-vs-winner breakdown (no visual linking which tournaments were hosted and also won by the host), so the claim that \"8 out of 22 tournaments (36.36%) resulted in the host nation winning\" and the listed specific host winners cannot be verified from these images alone. To evaluate the host-win relationship directly, a chart or table marking each tournament with host and winner (and indicating cases where they coincide) is needed.",280 "confidence": 4.0,281 "notebook": "fifa-world-cup-1930-2022.ipynb",282 "id": 237,283 "figure": "<image_id:14> <image_id:15> <image_id:16>",284 "dataset_size_mb": 0.01723575592041,285 "dataset": "iamsouravbanerjee/fifa-football-world-cup-dataset"286 },287 {288 "data_type": "tabular data",289 "domain": "Data Analysis",290 "task_type": "Data Preparation & Wrangling",291 "language": "Python",292 "question": "From the raw AIDS_Classification.csv data, split the features and target (infected) into a 70% training and 30% test set, then rebalance only the training set with SMOTE. What are the resulting shapes before and after SMOTE?",293 "reasoning": "Load the full dataset and separate predictors from the binary target variable. Partition the data into training and test subsets with a 70/30 split to create a held-out evaluation set. Examine the initial shapes to confirm the correct split. Since class imbalance can hinder model learning, apply SMOTE to the training set alone to synthetically upsample the minority class, ensuring the test set remains untouched for unbiased evaluation. Finally, verify the new training set size after resampling and confirm the test set shapes are unchanged.",294 "answer": "Before SMOTE: x_train (1497, 22), y_train (1497,), x_test (642, 22), y_test (642,). After SMOTE on the training set: x_train (2246, 22), y_train (2246,), x_test (642, 22), y_test (642,).",295 "confidence": 4.0,296 "notebook": "aids-classification-eda-acc-91.ipynb",297 "id": 299,298 "figure": null,299 "dataset_size_mb": 4.685982704162598,300 "dataset": "aadarshvelu/aids-virus-infection-prediction"301 },302 {303 "data_type": "tabular data",304 "domain": "Data Analysis",305 "task_type": "Data Preparation & Wrangling",306 "language": "Python",307 "question": "Given the stroke prediction dataset (healthcare-dataset-stroke-data.csv), impute the missing BMI values by training a decision tree regressor using age and encoded gender, fill the missing BMI entries, and report how many missing values remain in the dataset.",308 "reasoning": "Start by loading the dataset and inspecting missing values to identify that BMI has gaps. Encode gender numerically to be used by a regression model. Split the data into records with known BMI and those with missing BMI. Train a decision tree regressor using age and encoded gender on records that have BMI available. Use the trained model to predict BMI for records where it is missing. Fill these predictions back into the dataset. Finally, re-check the dataset for any remaining missing values to ensure the imputation was successful.",309 "answer": "Missing values: 0",310 "confidence": 4.0,311 "notebook": "predicting-a-stroke-shap-lime-explainer-eli5.ipynb",312 "id": 312,313 "figure": null,314 "dataset_size_mb": 0.302287101745605,315 "dataset": "fedesoriano/stroke-prediction-dataset"316 },317 {318 "data_type": "tabular data",319 "domain": "Data Analysis",320 "task_type": "Model Evaluation & Selection",321 "language": "Python",322 "question": "With the stroke prediction dataset, compute the class distribution for the target (stroke) and quantify the null accuracy (always predicting the majority class) and its inverse as a baseline for model evaluation.",323 "reasoning": "Load the raw dataset and examine the target variable to determine the counts of positive (stroke) and negative (no stroke) cases. The null accuracy is the proportion of the majority class in the dataset, while the inverse is the proportion of the minority class. These quantities provide a baseline to assess whether the models perform better than naive predictions.",324 "answer": "Inverse of Null Accuracy: 0.0487279843444227; Null Accuracy: 0.9512720156555773",325 "confidence": 4.0,326 "notebook": "predicting-a-stroke-shap-lime-explainer-eli5.ipynb",327 "id": 313,328 "figure": null,329 "dataset_size_mb": 0.302287101745605,330 "dataset": "fedesoriano/stroke-prediction-dataset"331 },332 {333 "data_type": "tabular data",334 "domain": "Data Analysis",335 "task_type": "Data Preparation & Wrangling, Model Evaluation & Selection, Model Training & Optimization",336 "language": "Python",337 "question": "Using the stroke dataset after imputing BMI and encoding categorical features, balance the training data with SMOTE, then compare Random Forest, SVM, and Logistic Regression via 10-fold cross-validation on the training set using F1 score. Which model performs best and what are the mean F1 scores?",338 "reasoning": "Load the dataset and complete preprocessing by imputing missing BMI and encoding categorical variables. Split the data into training and testing subsets. Because the target classes are imbalanced, apply SMOTE to the training data to create a balanced training set. Set up pipelines for each algorithm (Random Forest, SVM, Logistic Regression), including scaling where appropriate, and run 10-fold cross-validation on the resampled training data using F1 score as the evaluation metric. Aggregate the mean F1 scores across the folds for each model and select the best-performing model based on the mean F1.",339 "answer": "Mean f1 scores: Random Forest mean: 0.9342717632419655; SVM mean: 0.8752026263943018; Logistic Regression mean: 0.8225682495045643. Best model: Random Forest.",340 "confidence": 4.0,341 "notebook": "predicting-a-stroke-shap-lime-explainer-eli5.ipynb",342 "id": 314,343 "figure": null,344 "dataset_size_mb": 0.302287101745605,345 "dataset": "fedesoriano/stroke-prediction-dataset"346 },347 {348 "data_type": "tabular data",349 "domain": "Data Analysis",350 "task_type": "Model Evaluation & Selection, Model Training & Optimization",351 "language": "Python",352 "question": "Train and evaluate tuned models on the stroke dataset: after imputing BMI, encoding categorical variables, and balancing the training set with SMOTE, tune Logistic Regression (C=0.1, l2) and evaluate on the held-out test set. Report accuracy, recall for the stroke class, and F1 score, and determine whether it outperforms Random Forest and SVM on F1.",353 "reasoning": "From the raw data, first impute missing BMI values and encode categorical features. Split into train and test sets. Apply SMOTE on the training set to handle class imbalance. Tune Logistic Regression to use l2 regularization with C=0.1. Fit the tuned Logistic Regression on the resampled training data, then evaluate on the test set. Compute accuracy, recall for the positive class (stroke), and F1 score. Compare the F1 score with those of Random Forest and SVM (trained similarly) to determine which model performs best on the test set.",354 "answer": "Tuned Logistic Regression on test set: Accuracy Score: 0.7581772435001398; Recall (stroke class): 0.60; F1 Score: 0.19234360410831. This F1 outperforms Random Forest (0.16226415094339622) and SVM (0.15682281059063136) on the test set.",355 "confidence": 3.0,356 "notebook": "predicting-a-stroke-shap-lime-explainer-eli5.ipynb",357 "id": 315,358 "figure": null,359 "dataset_size_mb": 0.302287101745605,360 "dataset": "fedesoriano/stroke-prediction-dataset"361 },362 {363 "data_type": "tabular data",364 "domain": "Data Analysis",365 "task_type": "Model Evaluation & Selection, Model Training & Optimization",366 "language": "Python",367 "question": "For the tuned Logistic Regression model trained on the SMOTE-balanced data, adjust the decision threshold to 0.3 on the test set and report the confusion matrix, accuracy, F1 score, sensitivity (recall), and specificity.",368 "reasoning": "Start from the raw data and complete preprocessing (impute BMI, encode categoricals), split into train and test sets, and balance the training data with SMOTE. Fit the tuned Logistic Regression model. Instead of using the default 0.5 threshold, compute predicted probabilities on the test set and binarize predictions at a 0.3 threshold. Calculate the confusion matrix and derive accuracy, F1 score, sensitivity (true positive rate) and specificity (true negative rate) from it.",369 "answer": "With 0.3 threshold the Confusion Matrix is [[2133 1271], [33 140]]; Accuracy score: 0.6354487000279564; F1 score: 0.17676767676767677; Sensitivity: 0.8092485549132948; Specificity: 0.6266157461809636.",370 "confidence": 4.0,371 "notebook": "predicting-a-stroke-shap-lime-explainer-eli5.ipynb",372 "id": 316,373 "figure": null,374 "dataset_size_mb": 0.302287101745605,375 "dataset": "fedesoriano/stroke-prediction-dataset"376 },377 {378 "data_type": "tabular data",379 "domain": "Data Analysis",380 "task_type": "Data Preparation & Wrangling, Model Training & Optimization",381 "language": "Python",382 "question": "Given the student study hours dataset, how should the data be split for training and testing a regression model, and what are the dimensions of these splits?",383 "reasoning": "The dataset must be divided into training and testing sets to properly evaluate model performance. First, identify the features (independent variables) and target (dependent variable) in the dataset. Then, use a standard approach like train-test split with a specified proportion for testing. The training set should be large enough to allow the model to learn patterns, while the test set should be sufficient to provide a reliable evaluation. Calculate the dimensions of both sets to understand their composition and ensure the split is appropriate for the dataset size.",384 "answer": "The data was split using a test size of 0.20 (20% for testing), resulting in a training set with 20 samples (x_train and y_train with shape (20,1) and (20,)) and a test set with 5 samples (x_test and y_test with shape (5,1) and (5,)). This represents a standard 80-20 split that provides sufficient data for model training while maintaining a representative test set for evaluation.",385 "confidence": 4.0,386 "notebook": "student-study-hours-linear-regression.ipynb",387 "id": 331,388 "figure": null,389 "dataset_size_mb": 0.00092601776123,390 "dataset": "himanshunakrani/student-study-hours"391 },392 {393 "data_type": "tabular data",394 "domain": "Data Analysis",395 "task_type": "Data Preparation & Wrangling",396 "language": "Python",397 "question": "Given the raw CSV (breast-cancer-wisconsin data), clean the data by renaming columns, removing duplicate rows, treating '?' in 'Bare Nuclei' as missing, attempting mean imputation, dropping remaining missing rows, and converting 'Bare Nuclei' to integer. How many duplicates are there, what is the row count after removing duplicates, how many 'Bare Nuclei' values are missing after replacing '?' with NaN, what is the final row count after dropping missing rows, and what is the final dtype of 'Bare Nuclei'?",398 "reasoning": "Load the CSV and assign descriptive column names. Check for and count duplicate rows to understand redundancy, then remove them to ensure unique samples. Recognize that 'Bare Nuclei' contains '?' placeholders and convert these to missing values to enable proper numeric handling. Inspect missingness specifically in 'Bare Nuclei' after the conversion to quantify gaps. Attempt mean imputation from the available numeric values, then verify if any missing values remain. Drop any remaining missing rows to ensure a fully numeric dataset for later modeling. Finally, coerce 'Bare Nuclei' to integer type and validate the resulting data types.",399 "answer": "Duplicates: 8. Rows after removing duplicates: 690. Missing 'Bare Nuclei' after replacing '?' with NaN: 16. Final rows after dropping remaining missing values: 674. Final dtype of 'Bare Nuclei': int64.",400 "confidence": 4.0,401 "notebook": "breast-cancer-svm.ipynb",402 "id": 345,403 "figure": null,404 "dataset_size_mb": 0.018967628479003,405 "dataset": "saurabhbadole/breast-cancer-wisconsin-state"406 },407 {408 "data_type": "tabular data",409 "domain": "Data Analysis",410 "task_type": "Statistical Testing & Inference",411 "language": "Python",412 "question": "Using the cleaned breast cancer dataset, compute Pearson correlations between all features and the target 'Class'. Which five features have the highest positive correlation with 'Class', and what are their correlation coefficients?",413 "reasoning": "Start from the cleaned numeric dataset. Compute the full Pearson correlation matrix to quantify linear associations between features and the target. Extract the column corresponding to 'Class' and sort feature correlations in descending order to identify the most predictive signals. Select the top five positively correlated features and report their exact coefficients.",414 "answer": "Top five positively correlated features with 'Class': 1) Uniformity of Cell Shape: 0.820543, 2) Uniformity of Cell Size: 0.820527, 3) Bare Nuclei: 0.820397, 4) Bland Chromatin: 0.758376, 5) Normal Nucleoli: 0.721841.",415 "confidence": 4.0,416 "notebook": "breast-cancer-svm.ipynb",417 "id": 346,418 "figure": null,419 "dataset_size_mb": 0.018967628479003,420 "dataset": "saurabhbadole/breast-cancer-wisconsin-state"421 },422 {423 "data_type": "tabular data",424 "domain": "Data Analysis",425 "task_type": "Feature Engineering & Preparation, Reporting & Interpretation",426 "language": "Python",427 "question": "Using the cleaned breast cancer dataset, drop 'Sample code number' and 'Uniformity of Cell Size' to reduce redundancy and then visualize the correlation heatmap of the remaining features. What key relationships are visible in the heatmap?",428 "reasoning": "Remove the identifier-like column to avoid spurious associations and drop one of the highly collinear uniformity measures to reduce redundancy. Recompute the correlation matrix on the reduced feature set. Visualize the correlations as a heatmap to spot strong positive or negative relationships and clusters. Summarize the prominent associations, especially those involving the target.",429 "answer": "The heatmap on the reduced feature set shows strong positive associations between 'Class' and features such as 'Uniformity of Cell Shape', 'Bare Nuclei', 'Bland Chromatin', 'Normal Nucleoli', and 'Clump Thickness'. Overall, a cluster of cytological features remains strongly inter-correlated and positively related to malignancy. <image_id:1>",430 "confidence": 4.0,431 "notebook": "breast-cancer-svm.ipynb",432 "id": 347,433 "figure": "<image_id:1>",434 "dataset_size_mb": 0.018967628479003,435 "dataset": "saurabhbadole/breast-cancer-wisconsin-state"436 },437 {438 "data_type": "tabular data",439 "domain": "Domain-Specific Applications",440 "task_type": "Model Evaluation & Selection, Prediction & Forecasting",441 "language": "Python",442 "question": "Given the NBA player statistics dataset from 1980-2022, how should you evaluate a model predicting All-Star selections when only 24 players are selected each season (12 per conference), and what are the predicted ranks for LeBron James and Luka Dončić for the 2023 All-Star selection?",443 "reasoning": "First, the dataset must be prepared with appropriate features that capture player performance and context. Then, the model must be trained using a custom evaluation metric that accounts for the constraint of only 24 All-Stars being selected each season. This involves ranking players within each conference and selecting the top 12. For prediction, the model calculates the probability of All-Star selection for each player, ranks them within their conference, and identifies the top 12 players. This approach ensures the model correctly handles the selection constraint rather than treating it as a simple binary classification problem.",444 "answer": "The appropriate evaluation metric is the proportion of correctly predicted All-Stars out of the 24 selected per season, considering the top 12 in each conference. The model achieves an average accuracy of 97.96% across multiple train/test splits. For the 2023 predictions: LeBron James is predicted to have rank 1 in the Western Conference, while Luka Dončić is predicted to have rank 2 in the Western Conference.",445 "confidence": 4.0,446 "notebook": "2023-nba-all-star-predictions.ipynb",447 "id": 356,448 "figure": null,449 "dataset_size_mb": 30.517166137695312,450 "dataset": "sumitrodatta/nba-aba-baa-stats"451 },452 {453 "data_type": "tabular data",454 "domain": "Data Analysis",455 "task_type": "Data Ingestion & Integration, Reporting & Interpretation",456 "language": "Python",457 "question": "I have the gender classification dataset ('Transformed Data Set - Sheet1.csv'). Starting from the raw CSV, report the dataset size, column data types and non-null counts, the number of unique categories per column (including the most frequent category and its frequency), and determine whether the gender classes are balanced. Also produce a gender countplot.",458 "reasoning": "Begin by loading the CSV into a tabular structure. Inspect column types and non-null counts to ensure data is fully populated. Retrieve the dataset shape to understand rows and columns. Summarize each categorical column to get the count of unique values and identify the most frequent category along with its frequency. Examine the target column distribution by counting occurrences of each gender to assess class balance. Finally, visualize the class distribution using a countplot to confirm balance visually.",459 "answer": "Shape: (66, 5). All columns are object type with 66 non-null entries each. Unique categories per column and top values from summary: Favorite Color: unique=3, top=Cool (freq=37); Favorite Music Genre: unique=7, top=Rock (freq=19); Favorite Beverage: unique=6, top=Doesn't drink (freq=14); Favorite Soft Drink: unique=4, top=Coca Cola/Pepsi (freq=32); Gender: unique=2, top=F (freq=33). Gender distribution is balanced: F=33, M=33. <image_id:0>",460 "confidence": 4.0,461 "notebook": "gender-classification.ipynb",462 "id": 440,463 "figure": "<image_id:0>",464 "dataset_size_mb": 0.0023059844970700002,465 "dataset": "hb20007/gender-classification"466 },467 {468 "data_type": "tabular data",469 "domain": "Data Analysis",470 "task_type": "Feature Engineering & Preparation, Reporting & Interpretation",471 "language": "Python",472 "question": "From the raw gender classification dataset ('Transformed Data Set - Sheet1.csv'), encode the 'Favorite Music Genre' into numeric labels and visualize the encoded distribution by gender. Also report the original unique genres, the set of encoded values observed, and the counts of each genre per gender.",473 "reasoning": "Load the dataset and inspect the unique categories in 'Favorite Music Genre' to understand the label space. Apply label encoding to map each genre category to an integer code. Confirm the encoded value set to ensure all categories are represented. Compute counts of each original genre within each gender to describe preferences. Visualize the distribution of the encoded genre values faceted by gender to show differences in patterns.",474 "answer": "Original unique genres: ['Rock', 'Hip hop', 'Folk/Traditional', 'Jazz/Blues', 'Pop', 'Electronic', 'R&B and soul']. After label encoding, the encoded values observed are [6, 2, 1, 3, 4, 0, 5]. Counts per genre by gender: F—Pop=13, Rock=10, Jazz/Blues=3, Electronic=2, Folk/Traditional=2, R&B and soul=2, Hip hop=1; M—Rock=9, Hip hop=7, Electronic=6, Pop=4, R&B and soul=4, Folk/Traditional=2, Jazz/Blues=1. Visualization: faceted histogram of encoded genres by gender <image_id:3>.",475 "confidence": 4.0,476 "notebook": "gender-classification.ipynb",477 "id": 442,478 "figure": "<image_id:3>",479 "dataset_size_mb": 0.0023059844970700002,480 "dataset": "hb20007/gender-classification"481 },482 {483 "data_type": "tabular data",484 "domain": "Data Analysis",485 "task_type": "Reporting & Interpretation, Statistical Testing & Inference",486 "language": "Python",487 "question": "Starting from the raw gender classification dataset ('Transformed Data Set - Sheet1.csv'), analyze beverage preferences: report the distribution of 'Favorite Beverage' across all users and the distribution of 'Favorite Soft Drink' within each gender. Include a pie chart for beverages and a grouped countplot for soft drinks by gender.",488 "reasoning": "Load the dataset and tally counts for the 'Favorite Beverage' column to understand overall drinking preferences. Visualize this distribution with a pie chart to show relative shares. Next, compute counts of 'Favorite Soft Drink' stratified by gender to characterize soft drink preferences within each class. Present these with a grouped countplot to facilitate comparison across genders.",489 "answer": "Favorite Beverage counts: Doesn't drink=14, Beer=13, Other=11, Wine=10, Vodka=9, Whiskey=9. Visualization: beverage pie chart <image_id:4>. Favorite Soft Drink by gender—F: Coca Cola/Pepsi=17, 7UP/Sprite=8, Fanta=6, Other=2; M: Coca Cola/Pepsi=15, Fanta=8, 7UP/Sprite=5, Other=5. Visualization: soft drink countplot by gender <image_id:5>.",490 "confidence": 4.0,491 "notebook": "gender-classification.ipynb",492 "id": 444,493 "figure": "<image_id:4> <image_id:5>",494 "dataset_size_mb": 0.0023059844970700002,495 "dataset": "hb20007/gender-classification"496 },497 {498 "data_type": "tabular data",499 "domain": "Data Analysis",500 "task_type": "Exploratory Data Analysis, Model Evaluation & Selection, Model Training & Optimization",501 "language": "Python",502 "question": "Given a loan approval dataset containing financial information of applicants, how does the credit score (cibil_score) influence the loan approval decision, and what is the threshold value that separates approved and rejected applications?",503 "reasoning": "First, the dataset would be loaded and cleaned to ensure data integrity. Then, a scatter plot would be created to visualize the relationship between credit scores and loan approval status. The distribution of credit scores for approved and rejected applications would be analyzed to identify patterns. Statistical analysis would be performed to determine the cutoff point where the majority of approved applications have scores above a certain threshold. This would involve examining the concentration of data points in the scatter plot and identifying where the two classes (approved/rejected) are most distinctly separated. Finally, the threshold value would be determined by identifying where the transition between approval and rejection occurs most clearly in the data.",504 "answer": "The credit score is highly related to loan approval status, with a clear separation point between 540-550. Applications with credit scores above this threshold have a significantly higher chance of being approved. Specifically, the threshold value where the separation becomes clear is around 540-550. While scores below 579 are classified as 'Poor', scores above 540-550 still have a good chance of approval, suggesting lenders have flexibility in their decision-making. The highest accuracy model (Random Forest) achieved 97.3% accuracy. Best accuracy: 97.3%.",505 "confidence": 4.0,506 "notebook": "loan-prediction-eda-x-2-anova-test-rf-97.ipynb",507 "id": 460,508 "figure": null,509 "dataset_size_mb": 0.366532325744628,510 "dataset": "architsharma01/loan-approval-prediction-dataset"511 },512 {513 "data_type": "tabular data",514 "domain": "Data Analysis",515 "task_type": "Statistical Testing & Inference",516 "language": "Python",517 "question": "Given a loan approval dataset with categorical variables including education level and employment status, does the education level significantly influence the loan approval status, and what statistical test was used to determine this relationship?",518 "reasoning": "First, the dataset would be loaded and examined to identify the categorical variables of interest. A contingency table would be created to show the distribution of loan approval status across different education levels. A Chi-Square test would then be performed to determine if there is a statistically significant relationship between education level and loan approval status. The Chi-Square test would calculate a test statistic and p-value, which would be compared to a significance level (typically 0.05) to determine if the null hypothesis of independence can be rejected. The expected frequencies would be calculated to understand the theoretical distribution under the null hypothesis. This process would allow for an objective assessment of whether education level affects loan approval decisions.",519 "answer": "The Chi-Square test was used to examine the relationship between education level and loan approval status. The test yielded a Chi-Square value of 0.08395754138250573 with a p-value of 0.7720042291016309. Since the p-value is greater than 0.05, we cannot reject the null hypothesis, indicating that there is no significant association between education level and loan approval status. The data does not provide evidence to conclude that education level affects loan approval decisions.",520 "confidence": 4.0,521 "notebook": "loan-prediction-eda-x-2-anova-test-rf-97.ipynb",522 "id": 464,523 "figure": null,524 "dataset_size_mb": 0.366532325744628,525 "dataset": "architsharma01/loan-approval-prediction-dataset"526 },527 {528 "data_type": "tabular data",529 "domain": "Data Analysis",530 "task_type": "Data Ingestion & Integration, Data Preparation & Wrangling",531 "language": "Python",532 "question": "I have three CSV files of Michelin restaurants (one-star, two-stars, and three-stars). After concatenating them, adding a 'Michelin Stars' column, dropping 'url' and 'zipCode', and removing rows with missing 'city', what is the final dataset shape and how many values are missing in the 'price' column?",533 "reasoning": "Start by loading the three raw CSVs. Concatenate them row-wise to create one unified dataset. Create a new column that encodes the number of Michelin stars per row (1, 2, or 3). Remove fields that are not useful predictors ('url' and 'zipCode'). Exclude any records where 'city' is missing to ensure clean geographic features. Finally, inspect the resulting table to confirm its shape and count the missing values for the 'price' feature.",534 "answer": "Final dataset shape: [693 rows x 9 columns]. Missing 'price' values: 176.",535 "confidence": 3.0,536 "notebook": "michelin-restaurants-eda-missing-price-prediction.ipynb",537 "id": 476,538 "figure": null,539 "dataset_size_mb": 0.10621452331542901,540 "dataset": "jackywang529/michelin-restaurants"541 },542 {543 "data_type": "text data",544 "domain": "Natural Language Processing",545 "task_type": "Feature Engineering & Preparation",546 "language": "Python",547 "question": "Using the restaurant 'name' text, create a bag-of-words representation and apply Truncated SVD to reduce it to 300 components. What proportion of variance is captured cumulatively by these 300 components?",548 "reasoning": "Begin with the raw text in the 'name' field. Convert the text into a count-based bag-of-words representation to obtain a high-dimensional sparse matrix. Apply a dimensionality reduction method (Truncated SVD) to project the sparse matrix into a lower-dimensional space of 300 components. Compute the cumulative explained variance across the components to quantify how much of the original signal is preserved. Report the cumulative proportion retained.",549 "answer": "Using 300 components explains more than 70% of the variance (cumulative explained variance > 0.70).",550 "confidence": 3.0,551 "notebook": "michelin-restaurants-eda-missing-price-prediction.ipynb",552 "id": 477,553 "figure": null,554 "dataset_size_mb": 0.10621452331542901,555 "dataset": "jackywang529/michelin-restaurants"556 },557 {558 "data_type": "tabular data",559 "domain": "Model Evaluation",560 "task_type": "Prediction & Forecasting, Reporting & Interpretation",561 "language": "Python",562 "question": "After training the Random Forest classifier on known entries, use it to impute the missing 'price' categories for restaurants with NaN 'price'. Show five sample predictions with key fields.",563 "reasoning": "Identify the subset of records where 'price' is missing. Prepare their features using the same engineering scheme as in training, including count-encoded categorical variables and the additional rough text-based prediction. Apply the trained Random Forest classifier to infer the price category for each missing entry. Present a small sample of the resulting predictions to validate that the pipeline produces reasonable outputs.",564 "answer": "Sample predictions:\n1) Driftwood | year: 2019 | latitude: 50.18890 | longitude: -4.97088 | city: 1 | region: 162 | cuisine: 108 | predicted price: 4.0 | Michelin Stars: 1\n2) Yauatcha Soho | year: 2019 | latitude: 51.51367 | longitude: -0.13520 | city: 3 | region: 162 | cuisine: 15 | predicted price: 3.0 | Michelin Stars: 1\n3) Martin Wishart | year: 2019 | latitude: 55.97552 | longitude: -3.17019 | city: 2 | region: 162 | cuisine: 108 | predicted price: 4.0 | Michelin Stars: 1\n4) Ledbury | year: 2019 | latitude: 51.51674 | longitude: -0.20007 | city: 2 | region: 162 | cuisine: 108 | predicted price: 4.0 | Michelin Stars: 2\n5) The Peat Inn | year: 2019 | latitude: 56.27861 | longitude: -2.88458 | city: 1 | region: 162 | cuisine: 15 | predicted price: 4.0 | Michelin Stars: 1",565 "confidence": 3.0,566 "notebook": "michelin-restaurants-eda-missing-price-prediction.ipynb",567 "id": 480,568 "figure": null,569 "dataset_size_mb": 0.10621452331542901,570 "dataset": "jackywang529/michelin-restaurants"571 },572 {573 "data_type": "tabular data",574 "domain": "Statistical Testing & Experimentation",575 "task_type": "Statistical Testing & Inference",576 "language": "Python",577 "question": "Using the body performance dataset with 13,393 individuals, conduct an independent t-test to compare the mean 'body fat_%' between male and female individuals. What is the t-statistic, p-value, and what conclusion can be drawn about whether there is a statistically significant difference in body fat percentage between genders?",578 "reasoning": "First, the dataset must be loaded and separated into two groups based on the 'gender' column: males and females. The 'body fat_%' values for each gender group must be extracted. An independent t-test should be performed to test the null hypothesis that the mean body fat percentage is equal between males and females, against the alternative hypothesis that they are different. The test produces a t-statistic and a corresponding p-value. The t-statistic measures how many standard errors the difference between group means is away from zero. The p-value represents the probability of observing such an extreme t-statistic if the null hypothesis were true. Using a standard significance level of α = 0.05, if the p-value is less than 0.05, the null hypothesis should be rejected, indicating a statistically significant difference. The effect size (Cohen's d) should also be examined to understand the practical significance of this difference.",579 "answer": "The independent t-test results show: Male Mean = 20.19%, Female Mean = 28.49%, t-statistic = -76.49, p-value = 0.0000 (essentially zero), Cohen's d = -1.37. Since the p-value is much less than 0.05, the null hypothesis is rejected. There is a statistically significant difference in body fat percentage between males and females. Females have a substantially higher mean body fat percentage (8.30 percentage points higher) compared to males. The large Cohen's d value of -1.37 indicates this is not only statistically significant but also practically significant, representing a very large effect size. This strong evidence suggests gender is an important factor in body fat percentage variation in this population.",580 "confidence": 4.0,581 "notebook": "guide-to-complete-statistical-analysis.ipynb",582 "id": 492,583 "figure": null,584 "dataset_size_mb": 0.726542472839355,585 "dataset": "kukuroo3/body-performance-data"586 },587 {588 "data_type": "tabular data",589 "domain": "Data Analysis",590 "task_type": "Data Ingestion & Integration, Pattern & Anomaly Detection, Reporting & Interpretation",591 "language": "Python",592 "question": "Using the body performance dataset, identify outliers in the 'systolic' blood pressure column using the Interquartile Range (IQR) method. What are the quartile values, IQR value, outlier thresholds, and how many values are identified as outliers? Visualize the distribution before and after outlier removal.",593 "reasoning": "First, the dataset must be loaded and the 'systolic' column extracted. The first quartile (Q1, 25th percentile) and third quartile (Q3, 75th percentile) must be calculated. The IQR is computed as Q3 minus Q1, representing the range containing the middle 50% of the data. The outlier thresholds are determined by: lower threshold = Q1 - 1.5 × IQR and upper threshold = Q3 + 1.5 × IQR. Any values falling below the lower threshold or above the upper threshold are classified as outliers. Data points beyond these thresholds are identified and counted. Boxplots should be created before outlier removal to visualize the original distribution with outliers displayed as individual points. After removing or replacing outliers (typically with 0 or NA), a second boxplot should be created to show the cleaned distribution. Comparing the two visualizations demonstrates the impact of outlier removal on the data distribution and reveals which values were considered outliers.",594 "answer": "For the 'systolic' column: Q1 = 120.0, Q3 = 141.0, IQR = 21.0, Lower threshold = Q1 - 1.5×IQR = 88.5, Upper threshold = Q3 + 1.5×IQR = 172.5. Values below 88.5 or above 172.5 are identified as outliers. The boxplot before removal shows several data points marked beyond the upper whisker, indicating systolic pressures above 172.5 mmHg are present in the dataset. After applying the IQR method to remove outliers by replacing them with 0, the boxplot shows a cleaner distribution with the upper outliers removed, resulting in a more compact visualization. The outlier removal process reveals that extreme systolic pressure values were present but represent a small proportion of the total 13,393 observations. <image_id:9> <image_id:10>",595 "confidence": 3.0,596 "notebook": "guide-to-complete-statistical-analysis.ipynb",597 "id": 494,598 "figure": "<image_id:9> <image_id:10>",599 "dataset_size_mb": 0.726542472839355,600 "dataset": "kukuroo3/body-performance-data"601 },602 {603 "data_type": "time series data",604 "domain": "Time Series",605 "task_type": "Data Preparation & Wrangling",606 "language": "Python",607 "question": "From the raw wind turbine dataset (Turbine_Data.csv), after parsing datetime and engineering time features, impute all numeric columns using median values and add missingness indicator columns for the imputed fields. After this preprocessing, how many missing values remain per column?",608 "reasoning": "Begin with the raw dataset, ensure the datetime is parsed and time features are engineered. Identify all numeric columns and create a binary indicator for each to record whether a value was originally missing. Impute each numeric column with the median to handle missingness robustly. Finally, recompute missing value counts across all columns to verify that imputation removed NA values.",609 "answer": "All columns report 0 missing values after imputation and adding missingness indicators:\nActivePower 0\nAmbientTemperatue 0\nBearingShaftTemperature 0\nBlade1PitchAngle 0\nBlade2PitchAngle 0\nBlade3PitchAngle 0\nControlBoxTemperature 0\nGearboxBearingTemperature 0\nGearboxOilTemperature 0\nGeneratorRPM 0\nGeneratorWinding1Temperature 0\nGeneratorWinding2Temperature 0\nHubTemperature 0\nMainBoxTemperature 0\nNacellePosition 0\nReactivePower 0\nRotorRPM 0\nTurbineStatus 0\nWTG 0\nWindDirection 0\nWindSpeed 0\nyear 0\nmonth 0\nday 0\nhour 0\nminute 0\nActivePower_is_missing 0\nAmbientTemperatue_is_missing 0\nBearingShaftTemperature_is_missing 0\nBlade1PitchAngle_is_missing 0\nBlade2PitchAngle_is_missing 0\nBlade3PitchAngle_is_missing 0\nControlBoxTemperature_is_missing 0\nGearboxBearingTemperature_is_missing 0\nGearboxOilTemperature_is_missing 0\nGeneratorRPM_is_missing 0\nGeneratorWinding1Temperature_is_missing 0\nGeneratorWinding2Temperature_is_missing 0\nHubTemperature_is_missing 0\nMainBoxTemperature_is_missing 0\nNacellePosition_is_missing 0\nReactivePower_is_missing 0\nRotorRPM_is_missing 0\nTurbineStatus_is_missing 0\nWindDirection_is_missing 0\nWindSpeed_is_missing 0",610 "confidence": 4.0,611 "notebook": "easy-wind-power-forecasting.ipynb",612 "id": 506,613 "figure": null,614 "dataset_size_mb": 21.947731018066406,615 "dataset": "theforcecoder/wind-power-forecasting"616 },617 {618 "data_type": "time series data",619 "domain": "Time Series",620 "task_type": "Statistical Testing & Inference",621 "language": "Python",622 "question": "Using the wind turbine ActivePower series Turbine_Data.csv, test for stationarity with the Augmented Dickey–Fuller (ADF) test. Report the test statistic, p-value, number of lags used, number of observations, and the conclusion about stationarity.",623 "reasoning": "Load the raw data, focus on the ActivePower time series, and apply the ADF unit root test. Extract the test statistic, p-value, the number of lags, and the number of observations used. Compare the p-value to a standard threshold (e.g., 0.05) to determine whether to reject the null hypothesis of a unit root and conclude stationarity or non-stationarity.",624 "answer": "ADF Test Statistic: -21.00166405825876\np-value: 0.0\n#Lags Used: 71\nNumber of Observations Used: 118152\nConclusion: strong evidence against the null hypothesis (unit root); reject H0. The series is stationary.",625 "confidence": 4.0,626 "notebook": "easy-wind-power-forecasting.ipynb",627 "id": 507,628 "figure": null,629 "dataset_size_mb": 21.947731018066406,630 "dataset": "theforcecoder/wind-power-forecasting"631 },632 {633 "data_type": "time series data",634 "domain": "Time Series",635 "task_type": "Model Evaluation & Selection, Model Training & Optimization, Prediction & Forecasting",636 "language": "Python",637 "question": "Train an ARIMA model on the first 5000 observations of the wind turbine ActivePower series from Turbine_Data.csv, then generate a 15-step forecast and evaluate it against a test window of the next 15 observations (index 1000 to 1014). Report MAPE, ME, MAE, MPE, RMSE, correlation, and minmax error.",638 "reasoning": "Start with the raw time series. Fit an ARIMA model using the first 5000 time-ordered observations to capture temporal dynamics. Define a test set of 15 subsequent points within the series and generate a 15-step-ahead forecast. Compare forecasted values to actuals using multiple metrics: MAPE, ME, MAE, MPE, RMSE, correlation, and minmax error to assess accuracy and bias.",639 "answer": "{'mape': 0.02618515791159242, 'me': -10.54358196555202, 'mae': 10.54358196555202, 'mpe': -0.02618515791159242, 'rmse': 11.626161629259185, 'corr': -5.4147287338992556e-15, 'minmax': 0.02618515791159237}",640 "confidence": 4.0,641 "notebook": "easy-wind-power-forecasting.ipynb",642 "id": 508,643 "figure": null,644 "dataset_size_mb": 21.947731018066406,645 "dataset": "theforcecoder/wind-power-forecasting"646 },647 {648 "data_type": "tabular data",649 "domain": "Domain-Specific Applications",650 "task_type": "Data Preparation & Wrangling",651 "language": "Python",652 "question": "Given the S&P 500 companies dataset (sp500_companies.csv), clean the data by dropping the 'State' column, filling the single missing revenue growth value using an external financial source (Yahoo Finance) for the affected ticker, and imputing missing full-time employee counts with the mode. After these steps, what is the updated revenue growth for CVS, and which tickers had missing 'Fulltimeemployees' and what value was used to impute them?",653 "reasoning": "Start by loading the tabular dataset to inspect its structure and identify missing values. Since 'State' is not required for the planned analysis and has multiple missing entries, remove this column to simplify the dataset. Next, locate the record with missing revenue growth; determine the ticker and fetch the correct figure from a reliable external source (Yahoo Finance) to replace the missing value, ensuring the dataset reflects realistic business metrics. Finally, identify entries with missing full-time employee counts. Because these are sparse and appear missing completely at random, and given that mean imputation would overestimate for the companies involved, impute with the mode of the 'Fulltimeemployees' column to maintain plausible counts. Report the updated revenue growth for the specific ticker and list the tickers that were imputed along with the imputation value.",654 "answer": "Updated revenue growth for CVS: 0.114. Tickers with missing 'Fulltimeemployees' were MET and CTXS, both imputed with 14000.0.",655 "confidence": 4.0,656 "notebook": "s-p-500-stock-analysis-for-beginners.ipynb",657 "id": 523,658 "figure": null,659 "dataset_size_mb": 92.66675853729248,660 "dataset": "andrewmvd/sp-500-stocks"661 },662 {663 "data_type": "tabular data",664 "domain": "Domain-Specific Applications",665 "task_type": "Data Ingestion & Integration, Data Preparation & Wrangling",666 "language": "Python",667 "question": "Using the S&P 500 companies dataset (sp500_companies.csv), assess missing EBITDA values within the Financial Services sector by comparing the number of missing entries to total companies per industry. Which industries in this sector have EBITDA missing for all their companies, and what are the counts?",668 "reasoning": "Load the dataset and isolate companies within the Financial Services sector. Count entries with missing EBITDA grouped by industry to quantify missingness. Separately count total companies per industry in the sector. Compare missing counts to total counts; industries where the missing count equals the total count are those with full missingness for EBITDA.",669 "answer": "EBITDA is missing for all companies in the following Financial Services industries: Banks—Diversified (4 missing of 4 total), Banks—Regional (14 missing of 14 total), and Insurance—Reinsurance (1 missing of 1 total).",670 "confidence": 4.0,671 "notebook": "s-p-500-stock-analysis-for-beginners.ipynb",672 "id": 527,673 "figure": null,674 "dataset_size_mb": 92.66675853729248,675 "dataset": "andrewmvd/sp-500-stocks"676 },677 {678 "data_type": "tabular data",679 "domain": "Business Analytics",680 "task_type": "Data Preparation & Wrangling, Reporting & Interpretation, Statistical Testing & Inference",681 "language": "Python",682 "question": "I have the dataset State_of_data_2022.csv. After excluding unemployed and students and defining five diversity groups (Homem branco, Homem negro, Mulher branca, Mulher negra, PCD) from the raw columns (Genero, Cor/raca/etnia, PCD), build a treemap of group representation and report the overall percentage of respondents who belong to any diversity group.",683 "reasoning": "Start by loading the raw CSV and standardizing column names. Derive a binary race variable to mark 'Negro' vs 'Não Negro' and exclude categories flagged for removal. Use gender, race, and disability to assign each respondent into one of five groups: Homem branco, Homem negro, Mulher branca, Mulher negra, or PCD. Create a binary indicator for 'diversidade' that is 'Sim' if the respondent is female, Black, or has a disability. Filter the dataset to include only employed respondents by removing unemployed and students and any with unspecified work status. Count the respondents per group and compute their percentages to feed a treemap. Finally, compute the share of respondents with 'diversidade' equal to 'Sim' to quantify overall representation.",684 "answer": "The treemap shows the distribution among the five groups (these rectangles sum to 100% of the plotted subset): Homem branco 52.28%, Homem negro 22.68%, Mulher branca 15.79%, Mulher negra 7.92%, PCD 1.33%. More than half of the plotted professionals are homens brancos (52.28%). The two smallest groups are PCD (1.33%) and mulher negra (7.92%). Note: the treemap displays proportions among these groups; it does not by itself state what share of the entire original sample belongs to any of these groups unless this plotted subset is the full sample after excluding unemployed and students. If you intend “belong to any diversity group” to mean “not Homem branco,” that share is 100% − 52.28% = 47.72%.",685 "confidence": 3.0,686 "notebook": "os-desafios-para-diversidade-em-dados.ipynb",687 "id": 537,688 "figure": "<image_id:0>",689 "dataset_size_mb": 9.347503662109375,690 "dataset": "datahackers/state-of-data-2022"691 },692 {693 "data_type": "tabular data",694 "domain": "Business Analytics",695 "task_type": "Data Preparation & Wrangling, Model Evaluation & Selection",696 "language": "Python",697 "question": "Using State_of_data_2022.csv, compute the percentage of PCD among employed respondents (based on the raw PCD column) and compare it with the IBGE benchmark of 6.7%. How many times smaller is the PCD share in the data-profession sample compared to Brazil?",698 "reasoning": "Load the raw dataset and normalize column names. Build the same employed-only cohort by excluding unemployed and students to align with the workforce view. From this cohort, compute the proportion of respondents with PCD marked as 'Sim'. Use the IBGE benchmark for people with disabilities (6.7%) as a reference. Compare the computed sample share to 6.7%, and calculate the relative factor indicating how many times smaller the sample share is.",699 "answer": "The PCD share among data professionals is 1.33% versus 6.7% in Brazil, which is about 5 times smaller.",700 "confidence": 2.0,701 "notebook": "os-desafios-para-diversidade-em-dados.ipynb",702 "id": 538,703 "figure": null,704 "dataset_size_mb": 9.347503662109375,705 "dataset": "datahackers/state-of-data-2022"706 },707 {708 "data_type": "tabular data",709 "domain": "Business Analytics",710 "task_type": "Data Preparation & Wrangling, Exploratory Data Analysis, Statistical Testing & Inference",711 "language": "Python",712 "question": "From State_of_data_2022.csv, after deriving the binary diversity indicator (Sim/Não) and recategorizing seniority such that Gestor? True maps to 'Gestor' and otherwise uses the 'Nivel' (Júnior, Pleno, Sênior), what are the seniority distributions for those in any diversity group versus those not in a diversity group, and what pattern emerges?",713 "reasoning": "Load the raw data, standardize column names, and construct the diversity grouping based on gender, race, and disability, along with the binary diversity flag. Recategorize seniority by converting managerial status to 'Gestor' and keeping 'Nivel' for non-managers. Restrict to employed respondents. Compute the column-normalized distribution of seniority levels within each of the two groups (diversidade 'Sim' vs 'Não'). Compare the proportions at each level to identify whether diversity-group respondents concentrate at junior or intermediate levels and whether non-diversity respondents are more evenly distributed, thus highlighting any 'scissor effect'.",714 "answer": "Among people in any diversity group, the distribution concentrates in Júnior (~30%) and Pleno (~30%), with only 17% in Gestor. Those not in a diversity group (homens brancos) are more evenly distributed across levels, around 20–25% in each, evidencing the 'efeito tesoura'.",715 "confidence": 2.0,716 "notebook": "os-desafios-para-diversidade-em-dados.ipynb",717 "id": 539,718 "figure": null,719 "dataset_size_mb": 9.347503662109375,720 "dataset": "datahackers/state-of-data-2022"721 },722 {723 "data_type": "tabular data",724 "domain": "Business Analytics",725 "task_type": "Data Preparation & Wrangling, Feature Engineering & Preparation, Statistical Testing & Inference",726 "language": "Python",727 "question": "Using State_of_data_2022.csv, map 'Faixa salarial' to ordered categories and bin them into three ranges (Até R$ 4k, R$ 4k–12k, Acima de R$ 12k). Within each diversity group, compute the share in each bin. What percentage of mulheres negras earn above R$ 12k, and how does this compare to homens brancos?",728 "reasoning": "Load the data and keep employed respondents. Harmonize the salary range labels into an ordered categorical scale. Group the detailed ranges into three bins reflecting up to 4k, between 4k and 12k, and above 12k per month. For each diversity group, compute the within-group percentage distribution across these bins, normalizing by group size so each group's shares sum to 100%. From these distributions, extract the share in the 'Acima de R$ 12k' bin for mulheres negras and for homens brancos to compare high-salary representation.",729 "answer": "Mulheres negras: 14% earn above R$ 12k, compared to 32% among homens brancos.",730 "confidence": 3.0,731 "notebook": "os-desafios-para-diversidade-em-dados.ipynb",732 "id": 540,733 "figure": null,734 "dataset_size_mb": 9.347503662109375,735 "dataset": "datahackers/state-of-data-2022"736 },737 {738 "data_type": "tabular data",739 "domain": "Business Analytics",740 "task_type": "Data Preparation & Wrangling",741 "language": "Python",742 "question": "Using the full State_of_data_2022.csv , flag respondents as 'Desempregado' if they report either unemployment status and compute the unemployment rate within each diversity group. Which group shows the highest unemployment and by how many percentage points do mulheres negras exceed homens brancos?",743 "reasoning": "Load the dataset and derive the five diversity groups from gender, race, and disability as defined. On the full sample, create a binary employment-status label where those reporting either unemployment option are marked 'Desempregado' and others 'Não desempregado'. Calculate the within-group unemployment rate for each diversity group by dividing the number of unemployed by the group size. Compare across groups to identify the highest unemployment. Compute the difference between the rates for mulheres negras and homens brancos.",744 "answer": "The highest unemployment rates are among women (negras and brancas), with mulheres negras exceeding homens brancos by 6 percentage points.",745 "confidence": 2.0,746 "notebook": "os-desafios-para-diversidade-em-dados.ipynb",747 "id": 541,748 "figure": null,749 "dataset_size_mb": 9.347503662109375,750 "dataset": "datahackers/state-of-data-2022"751 },752 {753 "data_type": "tabular data",754 "domain": "Data Analysis",755 "task_type": "Data Ingestion & Integration, Data Preparation & Wrangling",756 "language": "Python",757 "question": "I have the dataset at ev_charging_patterns.csv. Identify columns with missing values, then impute the median for Energy Consumed (kWh), Charging Rate (kW), and Distance Driven (since last charge) (km). What are the missing value counts before and after imputation?",758 "reasoning": "Load the raw CSV to inspect the structure and completeness of the data. Quantify missing values per column to determine where imputation is needed. Since the three numeric columns are suitable for median imputation and exhibit missing values, compute the medians from the available data and fill missing entries accordingly. After the imputation step, re-check missing counts to confirm that all previously missing values have been addressed and that no new missing values were introduced.",759 "answer": "Before imputation, missing values were: Energy Consumed (kWh): 66, Charging Rate (kW): 66, Distance Driven (since last charge) (km): 66. All other columns had 0 missing. After median imputation for those three columns, all columns have 0 missing.",760 "confidence": 3.0,761 "notebook": "ev-charging-eda.ipynb",762 "id": 554,763 "figure": null,764 "dataset_size_mb": 0.353842735290527,765 "dataset": "valakhorasani/electric-vehicle-charging-patterns"766 },767 {768 "data_type": "tabular data",769 "domain": "Data Analysis",770 "task_type": "Exploratory Data Analysis, Reporting & Interpretation",771 "language": "Python",772 "question": "Given the Superstore sales dataset with product categories and sub-categories spanning 2011-2014, how does sales performance vary across different product sub-categories, and what are the annual growth trends for each product line, identifying which categories are top performers and which require strategic attention?",773 "reasoning": "First, the preprocessed dataset should be aggregated at the sub-category level to calculate total sales for each product sub-category across the entire period. The data should be sorted by sales magnitude to identify the top and bottom performers. Next, the dataset should be aggregated by sub-category and year to track how sales evolved for each product line over the four-year period. This yearly breakdown allows visualization of growth trajectories and identification of products with consistent vs. volatile sales. For each sub-category, the annual growth rate should be calculated year-over-year to identify which products are growing fastest and slowest. The average annual growth rate (AAGR) across the period should be computed for all sub-categories to rank them by growth performance. Finally, visualizations should be created to compare sales across sub-categories by year and category type to highlight relative performance within each main category (Furniture, Technology, Office Supplies).",774 "answer": "The two charts do answer the question: the horizontal bar chart shows total sales by sub-category (colored by major category), and the grouped bar chart shows yearly (2011–2014) sales per sub-category so you can see annual trends. Key observations visible in the images: \n\n- Top performers (by total sales): Phones and Chairs are the clear leaders (both substantially higher than other sub-categories). Tables, Binders and Storage are mid-to-high performers, followed by Copiers and Accessories. \n\n- Low performers (by total sales): Fasteners, Envelopes, Art, and Labels are the smallest contributors and appear to require strategic attention. \n\n- Annual trends (2011–2014): many sub-categories dip from 2011→2012 (notably Binders, Phones, Storage, Supplies, and Tables), then recover in 2013 and 2014. 2013 and 2014 are generally stronger years for most sub-categories, with 2014 showing the largest single-year increases for several top lines (e.g., Phones and Chairs). \n\n- Strong growth patterns: Copiers and Accessories show steady year‑over‑year increases across 2011–2014. Phones and Chairs rebound after a 2012 dip and reach their highest sales in 2014. \n\n- Volatile/uneven patterns: Machines and some other sub-categories show fluctuation rather than steady growth (not consistently flat). \n\nOverall: concentrate on high-volume, fast-recovering sub-categories (Phones, Chairs, Tables, Storage) for growth opportunities, and investigate low-volume items (Fasteners, Envelopes, Art, Labels) for possible consolidation or different strategies. The original answer’s specific dollar amounts and some growth-rate rankings do not match the visual charts.",775 "confidence": 4.0,776 "notebook": "retail-sales-exploratory-data-analysis-eda.ipynb",777 "id": 569,778 "figure": "<image_id:7> <image_id:8>",779 "dataset_size_mb": 5.488334655761719,780 "dataset": "ishanshrivastava28/superstore-sales"781 },782 {783 "data_type": "tabular data",784 "domain": "Data Analysis",785 "task_type": "Exploratory Data Analysis, Reporting & Interpretation",786 "language": "Python",787 "question": "Given the Superstore sales data across four U.S. regions (Central, South, East, West) from 2011-2014, how does regional sales performance vary both temporally and by product category, what are the seasonal patterns within each region, and which regions demonstrate the strongest growth potential?",788 "reasoning": "First, the dataset should be filtered and aggregated by region and month across all years to identify seasonal patterns within each region. Line plots should be created for each year showing monthly sales trends by region to visualize how seasonal variations differ across regions. Next, the dataset should be aggregated by region and sub-category to analyze product-level performance within each geographic area, revealing which products perform well in specific regions. This regional product analysis helps identify regional preferences and market characteristics. Subsequently, the dataset should be aggregated by region and year to track how total sales evolved in each region across the four-year period. This allows calculation of year-over-year growth rates for each region. The average annual growth rate (AAGR) should be computed for each region to rank them by growth performance. Finally, the data should be examined to understand how different regions were affected by the overall sales dip in 2012, identifying which regions drove the decline and which maintained growth.",789 "answer": "Temporal / seasonal patterns: The monthly trend panels (2011–2014) show recurring high months in many years around September and November–December (holiday / back-to-school peaks), but patterns are not identical every year or across all regions—there are year-specific anomalies (for example, a large March 2011 spike in the South). Regional performance over time: The West has the largest annual totals and shows the clearest upward trajectory from 2011 to 2014; the East also grows steadily and is the second-largest region overall. Central has moderate totals with smaller increases, and the South has the weakest totals, including a pronounced dip in 2012 and a partial recovery thereafter. Product-category differences: The East and West dominate many high-revenue sub-categories (notably Phones, Chairs, Storage and other big-ticket / technology-like items), while Central and South tend to have lower totals across most sub-categories. The South shows relatively stronger sales than Central in some industrial/large-item categories (e.g., Machines), but it is not the leading region overall for most sub-categories. Growth potential: Based on yearly totals and the clear upward trend, West (strongest) and East (also strong) demonstrate the highest growth potential; Central is moderate, and South appears most vulnerable and in need of targeted improvement. Overall, the figures support region-level seasonality, category concentration in East/West for big-ticket items, and West/East as the primary growth opportunities.",790 "confidence": 4.0,791 "notebook": "retail-sales-exploratory-data-analysis-eda.ipynb",792 "id": 570,793 "figure": "<image_id:9> <image_id:10> <image_id:11>",794 "dataset_size_mb": 5.488334655761719,795 "dataset": "ishanshrivastava28/superstore-sales"796 },797 {798 "data_type": "tabular data",799 "domain": "Business Analytics",800 "task_type": "Exploratory Data Analysis, Model Evaluation & Selection, Reporting & Interpretation",801 "language": "Python",802 "question": "Given the Superstore sales dataset from 2011-2014 with product pricing, sales volumes, and profit data, how does profitability vary across product sub-categories, which products are operating at a loss, how significantly do discounts impact overall profit margins, and what is the relationship between discount levels and actual sales or profit performance?",803 "reasoning": "First, profit margin should be calculated for each product sub-category by dividing total net profit by total sales and multiplying by 100. Sub-categories should be ranked by profitability to identify the most and least profitable products. Products with negative profit margins should be flagged as operating at a loss. Second, profit before discount and profit after discount should be compared for each sub-category to quantify the discount impact. The percentage drop in profit due to discounts should be calculated for each sub-category. Third, the dataset should be segmented by discount level (0%, 10%, 20%, 40%, 50%, 60%, 70%, 80%) and summary statistics (mean, median, count) should be calculated for sales, selling price, and profit before discount for each discount group. This allows comparison of whether higher discounts are associated with higher sales or higher-value products. Fourth, the distribution of discounts should be visualized to understand what proportion of orders receive each discount level. Finally, yearly profit margin trends should be analyzed to assess overall company profitability evolution.",804 "answer": "Summary of what the images actually show:\n- Average profit margins by sub-category (visible in three horizontal bar charts): Office Supplies sub-categories Labels, Paper and Envelopes show the highest average profit margins (around the low-40% range). Fasteners and some Office Supplies/Technology items are moderately profitable (around ~20–30%). Copiers and Accessories are strong within Technology (Copiers ~30+%, Accessories ~20+%). Several sub-categories have negative average margins: Binders (large negative), Appliances (negative), Tables (largest negative in Furniture, ~-15%), Bookcases (negative), and Machines show a small negative margin.\n- Net profit before vs after discounts (paired bar charts): Almost every sub-category shows a reduction in net profit after discounts. Copiers remain the largest positive net profit after discounts, followed by Phones and Accessories; Paper and Binders are mid-positive. Significant negative net profit after discounts is visible for Tables (a large swing from a positive pre-discount net profit to a large negative post-discount net profit), Bookcases (small negative), and Supplies (small negative). Several categories drop substantially though remain positive (e.g., Phones, Chairs, Accessories).\n- Discount distribution (histogram): Discounts are concentrated at 0% and 20% (large spikes). Fewer orders have high discounts (40%–80%), which appear as much smaller bars.\n- Relationship between discounts and profit (as visible): The before-vs-after charts show that discounts materially reduce net profit for many sub-categories — in some cases flipping a positive pre-discount profit to a net loss (Tables being the clearest example). The plots do not, however, show per-order sales or median sales by discount bin, so no direct conclusions about median sales or exact order counts at each discount level can be read from these images alone.\n\nNote: The original answer includes many precise numeric counts, medians and percentage-change figures and a contradictory statement about which sub-category is the single highest net-profit; those exact numeric claims and the count/median statistics are not directly shown in the provided images and therefore cannot be confirmed from these plots.",805 "confidence": 4.0,806 "notebook": "retail-sales-exploratory-data-analysis-eda.ipynb",807 "id": 571,808 "figure": "<image_id:12> <image_id:13> <image_id:14> <image_id:15> <image_id:20>",809 "dataset_size_mb": 5.488334655761719,810 "dataset": "ishanshrivastava28/superstore-sales"811 },812 {813 "data_type": "tabular data",814 "domain": "Data Analysis",815 "task_type": "Exploratory Data Analysis, Reporting & Interpretation",816 "language": "Python",817 "question": "Given the Superstore sales dataset across product categories and regions from 2011-2014, how do profit trends evolve for key product lines (Chairs in Furniture, Copiers in Technology, and various Office Supplies), and which products demonstrate consistent profitability growth versus stagnation or decline?",818 "reasoning": "First, the dataset should be aggregated by sub-category and year to calculate profit before discount for each product line across the four years. For each major product category (Furniture, Technology, Office Supplies), the profit data should be filtered and visualized with year-over-year comparisons. The profit trajectory should be analyzed to identify whether products show consistent growth, stagnation, decline, or volatility. Products like Chairs, Copiers, and Machines should be examined in detail to identify distinct patterns. For Furniture category products, the profit trends should be visualized to compare Chairs (high profit generator) versus Tables and Bookcases (loss-making). For Technology, Copiers should be compared to Machines to show contrast between rapid growth and stagnation. For Office Supplies, multiple products should be analyzed to identify those with upward trends (Appliances, Binders, Papers, Storage) versus flat performance. The cumulative or year-by-year profit should be tracked to quantify the growth magnitude.",819 "answer": "Based on the three charts: • Chairs (Furniture) — overall upward trajectory from 2011 to 2014 but not strictly monotonic: there is a visible dip in 2012 followed by stronger increases in 2013–2014, so Chairs show general growth but with a 2012 setback rather than a smooth year‑to‑year rise. • Copiers (Technology) — clear, large and consistent growth across the period; Copiers exhibit the steepest profit increase of the Technology subcategories and end 2014 as the category’s top performer. • Office Supplies (selected subcategories) — Binders and Paper show strong, steady growth from 2011 through 2014 (they are the largest and fastest‑growing office‑supply lines). Appliances and Storage also trend upward, though less dramatically. Many smaller office items (Art, Envelopes, Fasteners, Labels, Supplies) remain relatively small in profit and show only modest changes or mild variability rather than strong sustained growth. • Contrasting stagnation/decline — within Furniture, Bookcases decline across the period; Tables and some Furniture lines are more volatile. In Technology, Machines show a decline by 2014 after earlier years of higher profit. Overall takeaway: Copiers, Binders, and Paper are the clearest consistent growth stories; Chairs trend upward overall but with a mid‑period dip; Bookcases and Machines are areas of decline or concern.",820 "confidence": 3.0,821 "notebook": "retail-sales-exploratory-data-analysis-eda.ipynb",822 "id": 572,823 "figure": "<image_id:17> <image_id:18> <image_id:19>",824 "dataset_size_mb": 5.488334655761719,825 "dataset": "ishanshrivastava28/superstore-sales"826 },827 {828 "data_type": "tabular data",829 "domain": "Data Analysis",830 "task_type": "Data Ingestion & Integration",831 "language": "Python",832 "question": "Starting from the raw FastFoodNutritionMenuV2.csv, perform a special-value audit across all numeric nutrition columns. For each column, report the counts of special placeholders found and the total missing values (Null + specials).",833 "reasoning": "Load the data and iterate over each numeric nutrition column to detect non-numeric placeholders (e.g., space, '<1', '<5', and stray strings). For each column, sum the count of these specials with the Null count to get total missing. This informs a robust preprocessing plan.",834 "answer": "Calories: space=14; Null=1; Total missing=15. Calories from Fat: space=12; Null=506; Total missing=518. Total Fat (g): space=12; Null=57; Total missing=69. Saturated Fat (g): '5.5 g'=1, space=12; Null=57; Total missing=70. Trans Fat (g): space=12; Null=57; Total missing=69. Cholesterol (mg): space=14, '<5'=14; Null=1; Total missing=29. Sodium (mg): space=14, '<1'=1; Null=1; Total missing=16. Carbs (g): space=12, '<1'=1; Null=57; Total missing=70. Fiber (g): space=12, '<1'=15; Null=57; Total missing=84. Sugars (g): space=14, '<1'=15; Null=1; Total missing=30. Protein (g): space=12; Null=57; Total missing=69. Weight Watchers Pnts: space=11; Null=261; Total missing=272.",835 "confidence": 4.0,836 "notebook": "fast-food-nutrition-eda-data-analysis.ipynb",837 "id": 599,838 "figure": null,839 "dataset_size_mb": 0.15717601776123002,840 "dataset": "joebeachcapital/fast-food"841 },842 {843 "data_type": "tabular data",844 "domain": "Data Analysis",845 "task_type": "Data Preparation & Wrangling",846 "language": "Python",847 "question": "Given the raw FastFoodNutritionMenuV2.csv data, drop columns with excessive missingness ('Weight Watchers Pnts' and 'Protein (g)'), then clean the remaining nutrition columns by replacing non-numeric placeholders (space, '<1', '5.5 g', '<5') and imputing Nulls with the numeric mean (computed from valid values). What means are used for imputation per column, and do any Nulls remain after cleaning?",848 "reasoning": "Load the data and remove columns with large missingness to avoid bias. For each remaining numeric column, identify special placeholders and exclude them while computing the mean from valid numeric entries. Fill Nulls with this mean; replace '<1' with 0 and other placeholders with the mean. Convert the column to numeric and verify that missing values are eliminated. Report the per-column means used and confirm zero Nulls.",849 "answer": "Dropped columns: 'Weight Watchers Pnts', 'Protein (g)'. Imputation means used: Calories=287.909; Calories from Fat=118.034; Total Fat (g)=11.706; Saturated Fat (g)=4.077; Trans Fat (g)=0.141; Cholesterol (mg)=40.742; Sodium (mg)=428.477; Carbs (g)=39.06; Fiber (g)=1.461; Sugars (g)=24.153. Post-cleaning Null counts: all remaining columns have 0 Nulls.",850 "confidence": 4.0,851 "notebook": "fast-food-nutrition-eda-data-analysis.ipynb",852 "id": 600,853 "figure": null,854 "dataset_size_mb": 0.15717601776123002,855 "dataset": "joebeachcapital/fast-food"856 },857 {858 "data_type": "tabular data",859 "domain": "Data Analysis",860 "task_type": "Statistical Testing & Inference",861 "language": "Python",862 "question": "From the raw FastFoodNutritionMenuV2.csv, compute the skewness and kurtosis of the Calories distribution after loading the data (prior to modeling). What are the values?",863 "reasoning": "Load the dataset and select the Calories values as-is. Use statistical functions to compute distribution shape metrics: skewness to assess asymmetry and kurtosis to gauge tail weight and peakedness. Report the numerical outputs to characterize the distribution shape.",864 "answer": "Skewness (Calories): 1.183499003502487. Kurtosis (Calories): 1.8187640432934105.",865 "confidence": 4.0,866 "notebook": "fast-food-nutrition-eda-data-analysis.ipynb",867 "id": 601,868 "figure": null,869 "dataset_size_mb": 0.15717601776123002,870 "dataset": "joebeachcapital/fast-food"871 },872 {873 "data_type": "tabular data",874 "domain": "Domain-Specific Applications",875 "task_type": "Data Ingestion & Integration",876 "language": "Python",877 "question": "I have the CSV 'Latest Covid-19 India Status.csv' with state-wise COVID-19 metrics. Starting from the raw file, report the dataset dimensions, the column names, data types with non-null counts, the number of duplicated rows, and summarize missing values (including a missing-value heatmap).",878 "reasoning": "Begin by loading the CSV into a tabular structure. Inspect the dataset's dimensions to understand sample and feature counts. List the column names to verify schema alignment with expectations. Use an information summary to confirm each column's data type and whether any entries are missing by checking non-null counts. Assess data quality further by counting duplicated rows. Quantify missingness per column to ensure completeness, then visualize the missingness pattern with a heatmap to confirm the absence or presence of gaps at a glance.",879 "answer": "Shape: (36, 8). Columns: ['State/UTs', 'Total Cases', 'Active', 'Discharged', 'Deaths', 'Active Ratio (%)', 'Discharge Ratio (%)', 'Death Ratio (%)']. Dtypes and non-null counts: 36 non-null for each column; State/UTs: object; Total Cases, Active, Discharged, Deaths: int64; Active Ratio (%), Discharge Ratio (%), Death Ratio (%): float64. Duplicated rows: 0. Missing values per column: all zeros (State/UTs: 0, Total Cases: 0, Active: 0, Discharged: 0, Deaths: 0, Active Ratio (%): 0, Discharge Ratio (%): 0, Death Ratio (%): 0). <image_id:0>",880 "confidence": 4.0,881 "notebook": "covid-19-india-eda-visualization-report.ipynb",882 "id": 725,883 "figure": "<image_id:0>",884 "dataset_size_mb": 0.0020704269409170003,885 "dataset": "anandhuh/latest-covid19-india-statewise-data"886 },887 {888 "data_type": "tabular data",889 "domain": "Domain-Specific Applications",890 "task_type": "Reporting & Interpretation, Statistical Testing & Inference",891 "language": "Python",892 "question": "Using the raw 'Latest Covid-19 India Status.csv' data, compute the correlation matrix among all numeric variables and visualize it. Which variable pairs show the strongest positive and negative correlations, and what are their correlation values?",893 "reasoning": "Load the dataset and focus on numeric columns. Compute the pairwise Pearson correlation coefficients to quantify linear relationships. Identify the highest positive correlations among variables and the strongest negative correlations, noting both the variable pairs and the correlation magnitudes. Visualize the matrix using a heatmap to contextualize strengths and directions.",894 "answer": "Notable correlations from the computed matrix: Total Cases vs Discharged: 0.999838 (strong positive), Discharged vs Deaths: 0.903537 (strong positive), Total Cases vs Deaths: 0.900046 (strong positive), Total Cases vs Active: 0.611215 (moderate positive). The strongest negative correlation is between Active Ratio (%) and Discharge Ratio (%): -0.977407. Additional selected correlations: Active vs Discharge Ratio (%): -0.251663, Active vs Death Ratio (%): -0.190999, Deaths vs Death Ratio (%): 0.293068. Correlation heatmap shown. <image_id:1>",895 "confidence": 4.0,896 "notebook": "covid-19-india-eda-visualization-report.ipynb",897 "id": 726,898 "figure": "<image_id:1>",899 "dataset_size_mb": 0.0020704269409170003,900 "dataset": "anandhuh/latest-covid19-india-statewise-data"901 },902 {903 "data_type": "tabular data",904 "domain": "Domain-Specific Applications",905 "task_type": "Data Preparation & Wrangling, Exploratory Data Analysis",906 "language": "Python",907 "question": "From the raw state-wise COVID-19 data, identify the states with the highest and lowest total cases and report their full metrics (Total Cases, Active, Discharged, Deaths, and the three ratios).",908 "reasoning": "Load the dataset, then find the maximum and minimum values in the Total Cases column. Filter the rows corresponding to these extrema to retrieve the complete set of associated metrics for those states. Present the exact records for both the highest and lowest total case counts.",909 "answer": "Highest Total Cases: Maharashtra — Total Cases: 6,464,876; Active: 54,763; Discharged: 6,272,800; Deaths: 137,313; Active Ratio (%): 0.85; Discharge Ratio (%): 97.03; Death Ratio (%): 2.12. Lowest Total Cases: Andaman and Nicobar — Total Cases: 7,566; Active: 6; Discharged: 7,431; Deaths: 129; Active Ratio (%): 0.08; Discharge Ratio (%): 98.22; Death Ratio (%): 1.7.",910 "confidence": 4.0,911 "notebook": "covid-19-india-eda-visualization-report.ipynb",912 "id": 727,913 "figure": null,914 "dataset_size_mb": 0.0020704269409170003,915 "dataset": "anandhuh/latest-covid19-india-statewise-data"916 },917 {918 "data_type": "tabular data",919 "domain": "Domain-Specific Applications",920 "task_type": "Exploratory Data Analysis, Reporting & Interpretation",921 "language": "Python",922 "question": "Using the raw dataset, rank states by Death Ratio (%) and report the top five states with their death ratios. Include a visualization of these top five.",923 "reasoning": "Load the data and sort by the Death Ratio (%) in descending order to rank states by mortality proportion. Select the top five entries and extract their state names and death ratios. Visualize the top five using a horizontal bar chart for clear comparison.",924 "answer": "Top 5 by Death Ratio (%): 1) Punjab — 2.74; 2) Uttarakhand — 2.15; 3) Maharashtra — 2.12; 4) Nagaland — 2.06; 5) Goa — 1.84. Visualization shown. <image_id:5>",925 "confidence": 4.0,926 "notebook": "covid-19-india-eda-visualization-report.ipynb",927 "id": 729,928 "figure": "<image_id:5>",929 "dataset_size_mb": 0.0020704269409170003,930 "dataset": "anandhuh/latest-covid19-india-statewise-data"931 },932 {933 "data_type": "tabular data, time series data",934 "domain": "Business Analytics",935 "task_type": "Data Preparation & Wrangling, Exploratory Data Analysis, Reporting & Interpretation",936 "language": "Python",937 "question": "From the Online Sales Data.csv, compute total units sold by product category and visualize the category-wise distribution. Which product category has the highest total units sold, and what are the totals for all categories?",938 "reasoning": "Load the raw dataset and ensure it is clean. Group the data by product category and sum the units sold in each category to quantify category performance. Plot a bar chart to visualize category-wise totals and compare the bars to identify the top category by units sold.",939 "answer": "Category totals (Units Sold): Clothing: 145, Sports: 88, Books: 114, Electronics: 66, Home Appliances: 59, Beauty Products: 46. The highest is Clothing with 145 units. <image_id:3>",940 "confidence": 3.0,941 "notebook": "online-sales.ipynb",942 "id": 736,943 "figure": "<image_id:3>",944 "dataset_size_mb": 0.020736694335937,945 "dataset": "shreyanshverma27/online-sales-dataset-popular-marketplace-data"946 },947 {948 "data_type": "tabular data, time series data",949 "domain": "Business Analytics",950 "task_type": "Data Preparation & Wrangling, Feature Engineering & Preparation, Reporting & Interpretation",951 "language": "Python",952 "question": "Using the Online Sales Data.csv, parse the transaction Date field to derive the month, then compute and visualize total units sold per month. Which month has the highest total units sold, and what are the monthly totals?",953 "reasoning": "Load the raw data and convert the Date column into a proper date type to enable time-based analysis. Derive the month component for each transaction. Group by month and sum the units sold to obtain monthly totals. Visualize the monthly totals with a bar chart and compare values to determine the peak month.",954 "answer": "Monthly totals (Units Sold): Jan: 68, Feb: 77, Mar: 82, Apr: 65, May: 60, Jun: 61, Jul: 53, Aug: 52. The highest is March with 82 units. <image_id:7>",955 "confidence": 3.0,956 "notebook": "online-sales.ipynb",957 "id": 737,958 "figure": "<image_id:7>",959 "dataset_size_mb": 0.020736694335937,960 "dataset": "shreyanshverma27/online-sales-dataset-popular-marketplace-data"961 },962 {963 "data_type": "tabular data, time series data",964 "domain": "Business Analytics",965 "task_type": "Data Preparation & Wrangling, Reporting & Interpretation",966 "language": "Python",967 "question": "Filter the Online Sales Data.csv to the Clothing category and analyze performance: identify the top three products by total units sold, compute total units sold by day of the week for Clothing (with a visualization), and summarize the region and payment method distribution for Clothing transactions.",968 "reasoning": "Load the raw dataset and subset it to entries where the product category is Clothing. Aggregate units sold by product name and rank to identify top performers. For temporal patterns, group by day of the week and sum units sold, then visualize this distribution to spot the strongest day. Finally, check the distribution of region and payment method within the Clothing subset to understand where and how purchases were made.",969 "answer": "The image is a bar chart of total Clothing units sold by day of the week. Values visible in the chart: Sunday 26, Tuesday 24, Monday 21, Friday 20, and Wednesday/Thursday/Saturday each 18. Sunday is the strongest day. The chart does not show product-level totals, region breakdowns, or payment method distributions, so the original answer's top-three products and region/payment claims cannot be verified from this image.",970 "confidence": 3.0,971 "notebook": "online-sales.ipynb",972 "id": 738,973 "figure": "<image_id:5>",974 "dataset_size_mb": 0.020736694335937,975 "dataset": "shreyanshverma27/online-sales-dataset-popular-marketplace-data"976 },977 {978 "data_type": "tabular data, time series data",979 "domain": "Business Analytics",980 "task_type": "Data Preparation & Wrangling, Exploratory Data Analysis",981 "language": "Python",982 "question": "Using the Online Sales Data.csv, compute total revenue by Product Category and Region to determine which category–region pair contributes the highest revenue. Report the totals for all pairs and identify the top pair.",983 "reasoning": "Load the raw dataset and ensure financial fields are available. Group the data by both product category and region to capture revenue contributions across market segments. Sum total revenue within each pair. Compare the resulting sums to find the highest contributing category–region combination and report all pair totals for context.",984 "answer": "Total Revenue by Product Category and Region: Electronics–North America: 34982.41; Home Appliances–Europe: 18646.16; Sports–Asia: 14326.52; Clothing–Asia: 8128.93; Beauty Products–Europe: 2621.90; Books–North America: 1861.93. The highest revenue pair is Electronics in North America with 34982.41.",985 "confidence": 3.0,986 "notebook": "online-sales.ipynb",987 "id": 739,988 "figure": null,989 "dataset_size_mb": 0.020736694335937,990 "dataset": "shreyanshverma27/online-sales-dataset-popular-marketplace-data"991 },992 {993 "data_type": "tabular data",994 "domain": "Data Analysis",995 "task_type": "Model Evaluation & Selection",996 "language": "Python",997 "question": "Starting from the raw loan_data.csv and after preprocessing and training a K-Nearest Neighbors classifier (k=3) on an 80/20 train-test split using the selected features, report the classification report (precision, recall, F1-score, support for each class) on the test set and visualize the confusion matrix.",998 "reasoning": "Load and preprocess the data to obtain a clean numeric feature set, then select the correlation-filtered features. Split into train and test sets (80/20). Train a KNN model with k=3 on the training data. Evaluate the trained model on the test set to compute precision, recall, and F1-score for each class and the overall accuracy. Finally, summarize misclassification patterns by constructing a confusion matrix and visualizing it as a heatmap.",999 "answer": "Classification Report:\n- Class 0: precision 0.92, recall 0.93, f1-score 0.92, support 6990\n- Class 1: precision 0.75, recall 0.70, f1-score 0.72, support 2010\n- Overall accuracy: 0.88 (macro avg: precision 0.83, recall 0.82, f1-score 0.82; weighted avg: precision 0.88, recall 0.88, f1-score 0.88). Confusion matrix heatmap rendered. <image_id:18>",1000 "confidence": 4.0,1001 "notebook": "loan-approval-classification-eda-ml.ipynb",1002 "id": 744,1003 "figure": "<image_id:18>",1004 "dataset_size_mb": 3.443696975708008,1005 "dataset": "taweilo/loan-approval-classification-data"1006 },1007 {1008 "data_type": "tabular data",1009 "domain": "Data Analysis",1010 "task_type": "Data Ingestion & Integration, Data Preparation & Wrangling",1011 "language": "Python",1012 "question": "I have the CSV file 'Oscars-demographics-DFE.csv'. Load it into a DataFrame. If a UnicodeDecodeError occurs, detect the file's character encoding and re-load the data so it can be previewed successfully.",1013 "reasoning": "Start by attempting to read the CSV using the default UTF-8 decoding. If a Unicode decoding error happens, inspect the raw bytes of the file and use an encoding detection tool to estimate the most likely encoding along with a confidence score. Then, re-read the CSV while explicitly specifying the detected encoding. Validate that the data has loaded correctly by previewing the first few rows and confirming the reported number of rows and columns in the preview output.",1014 "answer": "Initial load failed with: ERROR: UnicodeDecodeError: 'utf-8' codec can't decode byte 0xcc in position 1: invalid continuation byte. Detected encoding from raw bytes: {'encoding': 'ISO-8859-1', 'confidence': 0.7289274470020289, 'language': ''}. Re-loading with encoding='ISO-8859-1' succeeded. Preview shows the first 5 rows and indicates [5 rows x 27 columns], with columns such as _unit_id, _golden, _unit_state, _trusted_judgments, _last_judgment_at, birthplace, birthplace:confidence, date_of_birth, date_of_birth:confidence, race_ethnicity, award, biourl, birthplace_gold, date_of_birth_gold, movie, person, race_ethnicity_gold, religion_gold, sexual_orientation_gold, year_of_award_gold.",1015 "confidence": 4.0,1016 "notebook": "how-to-resolve-a-unicodedecodeerror-for-a-csv-file.ipynb",1017 "id": 754,1018 "figure": null,1019 "dataset_size_mb": 0.08579063415527301,1020 "dataset": "fmejia21/demographics-of-academy-awards-oscars-winners"1021 },1022 {1023 "data_type": "tabular data",1024 "domain": "Data Analysis",1025 "task_type": "Data Ingestion & Integration",1026 "language": "Python",1027 "question": "After successfully loading 'Oscars-demographics-DFE.csv' with the correct encoding, extract the first five birthplace values and their associated confidence scores to verify the content integrity.",1028 "reasoning": "Load the dataset using the detected encoding to avoid decoding errors. Focus on the columns that store birthplace information and its confidence score. Preview the first five records of these two fields to confirm both the values and the presence of confidence metadata.",1029 "answer": "First five birthplace and confidence pairs: (1) 'Chisinau, Moldova' — 1.0, (2) 'Glasgow, Scotland' — 1.0, (3) 'Chisinau, Moldova' — 1.0, (4) 'Chicago, Il' — 1.0, (5) 'Salt Lake City, Ut' — 1.0.",1030 "confidence": 4.0,1031 "notebook": "how-to-resolve-a-unicodedecodeerror-for-a-csv-file.ipynb",1032 "id": 755,1033 "figure": null,1034 "dataset_size_mb": 0.08579063415527301,1035 "dataset": "fmejia21/demographics-of-academy-awards-oscars-winners"1036 },1037 {1038 "data_type": "tabular data",1039 "domain": "Data Analysis",1040 "task_type": "Data Ingestion & Integration",1041 "language": "Python",1042 "question": "After loading the 'Oscars-demographics-DFE.csv' file with the detected encoding, assess whether the following columns contain missing values in the first five records: 'birthplace_gold', 'date_of_birth_gold', 'race_ethnicity_gold', 'religion_gold', 'sexual_orientation_gold', and 'year_of_award_gold'.",1043 "reasoning": "Load the dataset using the correct encoding. Examine the specified columns in the top rows to see if they contain filled values or missing entries. Report the presence of missing values as displayed in the preview.",1044 "answer": "For the first five rows, all of the specified columns show missing values (NaN): 'birthplace_gold', 'date_of_birth_gold', 'race_ethnicity_gold', 'religion_gold', 'sexual_orientation_gold', and 'year_of_award_gold' are all NaN in rows 0–4.",1045 "confidence": 4.0,1046 "notebook": "how-to-resolve-a-unicodedecodeerror-for-a-csv-file.ipynb",1047 "id": 757,1048 "figure": null,1049 "dataset_size_mb": 0.08579063415527301,1050 "dataset": "fmejia21/demographics-of-academy-awards-oscars-winners"1051 },1052 {1053 "data_type": "tabular data",1054 "domain": "Data Analysis",1055 "task_type": "Data Ingestion & Integration",1056 "language": "Python",1057 "question": "I have the Among Us dataset stored under among-us-dataset. Before loading, list all available CSV files so I know which user files are present.",1058 "reasoning": "To verify the raw data sources, the directory containing the dataset should be recursively scanned to enumerate all files. This confirms the presence and names of each user CSV before any loading or processing. Listing the paths provides a reliable inventory of inputs and ensures downstream steps (like merging) are based on the correct files.",1059 "answer": "User21.csv\nUser10.csv\nUser6.csv\nUser4.csv\nUser2.csv\nUser8.csv\nUser1.csv\nUser17.csv\nUser14.csv\nUser19.csv\nUser7.csv\nUser20.csv\nUser3.csv\nUser22.csv\nUser11.csv\nUser18.csv\nUser9.csv\nUser5.csv\nUser13.csv\nUser25.csv\nUser23.csv\nUser12.csv\nUser24.csv\nUser16.csv\nUser15.csv",1060 "confidence": 4.0,1061 "notebook": "among-us-starter-notebook-and-eda.ipynb",1062 "id": 780,1063 "figure": null,1064 "dataset_size_mb": 0.19321250915527302,1065 "dataset": "ruchi798/among-us-dataset"1066 },1067 {1068 "data_type": "tabular data",1069 "domain": "Data Analysis",1070 "task_type": "Data Preparation & Wrangling, Feature Engineering & Preparation",1071 "language": "Python",1072 "question": "Given multiple Among Us user CSV files, merge them, split the 'Region/Game Code' field into separate 'Region' and 'Game Code' columns, parse 'Game Completed Date' into 'Game Date' and 24-hour 'Game Time', and convert duration columns to minutes. What are the 'Region', 'Game Code', 'Game Date', and 'Game Time' values for the first five processed records?",1073 "reasoning": "Start by loading and concatenating all user CSVs into a single table while adding a user identifier to retain provenance. Next, split the combined 'Region/Game Code' text field into two distinct columns for region and code to normalize server and match identifiers. Since all records include a completion timestamp, parse this into a date part and a 24-hour time part for temporal analysis, dropping the original combined string thereafter. Convert time-like fields measured as minutes and seconds into a consistent numeric minutes format for quantitative use. Finally, inspect the first few records to verify the transformations and report the requested columns.",1074 "answer": "For the first five records after processing:\n- Row 0: Region = Europe, Game Code = BKINIF, Game Date = 12/25/2020, Game Time = 21:18:14\n- Row 1: Region = Europe, Game Code = BKINIF, Game Date = 12/25/2020, Game Time = 21:07:12\n- Row 2: Region = Europe, Game Code = BKINIF, Game Date = 12/25/2020, Game Time = 20:54:11\n- Row 3: Region = Europe, Game Code = BKINIF, Game Date = 12/25/2020, Game Time = 20:44:21\n- Row 4: Region = Europe, Game Code = BKINIF, Game Date = 12/25/2020, Game Time = 20:34:38",1075 "confidence": 4.0,1076 "notebook": "among-us-starter-notebook-and-eda.ipynb",1077 "id": 781,1078 "figure": null,1079 "dataset_size_mb": 0.19321250915527302,1080 "dataset": "ruchi798/among-us-dataset"1081 },1082 {1083 "data_type": "tabular data",1084 "domain": "Data Analysis",1085 "task_type": "Data Ingestion & Integration, Feature Engineering & Preparation",1086 "language": "Python",1087 "question": "From the raw 'Game Completed Date' strings across all Among Us CSVs, identify the unique time zones present and decide whether to retain the time zone as a feature.",1088 "reasoning": "To determine if the time zone provides discriminative information, parse the trailing time zone substring from each completion timestamp and collect the set of unique values across all records. If the set contains only one value, the feature lacks variability and can be dropped as it does not contribute to analysis or modeling.",1089 "answer": "Unique time zones found: ['EST']. Since only one time zone is present, the time zone component was dropped.",1090 "confidence": 4.0,1091 "notebook": "among-us-starter-notebook-and-eda.ipynb",1092 "id": 782,1093 "figure": null,1094 "dataset_size_mb": 0.19321250915527302,1095 "dataset": "ruchi798/among-us-dataset"1096 },1097 {1098 "data_type": "tabular data",1099 "domain": "Data Analysis",1100 "task_type": "Data Preparation & Wrangling, Reporting & Interpretation",1101 "language": "Python",1102 "question": "After converting the Among Us 'Game Length' field from 'mm ss' strings to minutes as floating-point numbers, what are the first five values, and what are their associated Team, Outcome, and Ejected statuses?",1103 "reasoning": "First, ensure the duration strings are normalized by parsing minutes and seconds and converting them into a single numeric minutes value for each record. After transformation, verify correctness by inspecting the first few rows and pairing the converted 'Game Length' with key categorical context such as the player's team, the match outcome, and whether the player was ejected.",1104 "answer": "First five records after conversion:\n- Row 0: Team = Crewmate, Outcome = Loss, Ejected = Yes, Game Length = 9.60\n- Row 1: Team = Imposter, Outcome = Loss, Ejected = Yes, Game Length = 11.42\n- Row 2: Team = Crewmate, Outcome = Loss, Ejected = Yes, Game Length = 9.57\n- Row 3: Team = Crewmate, Outcome = Win, Ejected = No, Game Length = 7.50\n- Row 4: Team = Crewmate, Outcome = Loss, Ejected = No, Game Length = 20.37",1105 "confidence": 4.0,1106 "notebook": "among-us-starter-notebook-and-eda.ipynb",1107 "id": 783,1108 "figure": null,1109 "dataset_size_mb": 0.19321250915527302,1110 "dataset": "ruchi798/among-us-dataset"1111 },1112 {1113 "data_type": "tabular data",1114 "domain": "Pattern Mining & Association",1115 "task_type": "Data Preparation & Wrangling",1116 "language": "Python",1117 "question": "Given the groceries dataset with transaction records where each row represents a transaction with multiple items, how should the data be transformed into a format suitable for the Apriori algorithm?",1118 "reasoning": "First, we need to remove the transaction count column (first column) and for each transaction, collect all non-null items from the remaining columns into a single list. This transformed structure will be used as input for the Apriori algorithm. The process involves converting the dataset into a list of item lists where each inner list represents the items purchased in a single transaction.",1119 "answer": "The data should be transformed by creating a list of transactions where each transaction is a list of items (excluding the first column and any null values). For example, the first transaction with 4 items would be transformed to: ['citrus fruit', 'semi-finished bread', 'margarine', 'ready soups']",1120 "confidence": 4.0,1121 "notebook": "apriori-algorithm-on-grocery-market-data.ipynb",1122 "id": 789,1123 "figure": null,1124 "dataset_size_mb": 1.2527151107788081,1125 "dataset": "irfanasrullah/groceries"1126 },1127 {1128 "data_type": "tabular data",1129 "domain": "Pattern Mining & Association",1130 "task_type": "Model Training & Optimization",1131 "language": "Python",1132 "question": "Given the groceries dataset with 9835 transactions, what is the appropriate value for min_support in the Apriori algorithm, and how is it calculated?",1133 "reasoning": "First, we need to determine what constitutes a meaningful association. The min_support threshold should be set based on business context and how often an item or itemset appears. It's often calculated based on how many times a rule should appear to be considered significant. For example, if we want rules that appear at least 3 times a day for a 7-day week, we can calculate min_support as (3*7)/9835.",1134 "answer": "The appropriate min_support value is 0.0022, which is calculated as (3 times a day * 7 days) / 9835 total transactions. This ensures that only rules appearing frequently enough to be meaningful are considered.",1135 "confidence": 4.0,1136 "notebook": "apriori-algorithm-on-grocery-market-data.ipynb",1137 "id": 790,1138 "figure": null,1139 "dataset_size_mb": 1.2527151107788081,1140 "dataset": "irfanasrullah/groceries"1141 },1142 {1143 "data_type": "tabular data",1144 "domain": "Pattern Mining & Association",1145 "task_type": "Reporting & Interpretation",1146 "language": "Python",1147 "question": "Given the association rules extracted from the groceries dataset, what does a rule with support 0.005, confidence 0.50, and lift 4.2 mean, and how does it help retailers?",1148 "reasoning": "First, we need to understand what each metric represents. Support indicates the proportion of transactions containing the itemset. Confidence measures how often the consequent item is purchased when the antecedent item is purchased. Lift measures how much more likely the association is compared to what would be expected if the items were independent. A lift of 4.2 means the items are 4.2 times more likely to be purchased together than if they were independent.",1149 "answer": "This rule indicates that the item combination appears in 0.5% of transactions (support), and when the antecedent item is purchased, the consequent item is purchased 50% of the time (confidence). The lift of 4.2 shows that the items are 4.2 times more likely to be purchased together than if they were independent. This helps retailers understand which items are frequently bought together, which can inform product placement and cross-selling strategies.",1150 "confidence": 2.0,1151 "notebook": "apriori-algorithm-on-grocery-market-data.ipynb",1152 "id": 791,1153 "figure": null,1154 "dataset_size_mb": 1.2527151107788081,1155 "dataset": "irfanasrullah/groceries"1156 },1157 {1158 "data_type": "tabular data",1159 "domain": "Data Analysis",1160 "task_type": "Data Ingestion & Integration, Data Preparation & Wrangling",1161 "language": "Python",1162 "question": "Using the dataset 'dataset_olympics.csv', fill missing values by imputing the mean for 'Age', 'Height', and 'Weight' and replacing missing 'Medal' values with 0. Then, report the non-null counts and data types for all columns after cleaning.",1163 "reasoning": "Load the raw dataset and address missing values to ensure completeness: set the medal field to a default value (0) when missing and impute the numerical fields (Age, Height, Weight) with their respective column means. After imputation, validate the result by inspecting the schema, confirming that all columns have full non-null counts and noting the data types each column now holds.",1164 "answer": "Post-cleaning schema: 70000 entries, 15 columns. Non-null counts and dtypes—ID: 70000 non-null (int64); Name: 70000 non-null (object); Sex: 70000 non-null (object); Age: 70000 non-null (object); Height: 70000 non-null (object); Weight: 70000 non-null (object); Team: 70000 non-null (object); NOC: 70000 non-null (object); Games: 70000 non-null (object); Year: 70000 non-null (int64); Season: 70000 non-null (object); City: 70000 non-null (object); Sport: 70000 non-null (object); Event: 70000 non-null (object); Medal: 70000 non-null (object).",1165 "confidence": 4.0,1166 "notebook": "olympic-data-analysis.ipynb",1167 "id": 807,1168 "figure": null,1169 "dataset_size_mb": 8.969584465026855,1170 "dataset": "bhanupratapbiswas/olympic-data"1171 },1172 {1173 "data_type": "tabular data",1174 "domain": "Data Analysis",1175 "task_type": "Data Ingestion & Integration",1176 "language": "Python",1177 "question": "From the raw 'dataset_olympics.csv', what are all the distinct sports represented in the 'Sport' column?",1178 "reasoning": "After loading the dataset, examine the 'Sport' field to understand category coverage by extracting the set of distinct values. Listing the unique categories provides a clear view of the breadth of sports included in the dataset and is essential for downstream analyses such as grouping and stratification.",1179 "answer": "Basketball; Judo; Football; Tug-Of-War; Speed Skating; Cross Country Skiing; Athletics; Ice Hockey; Swimming; Badminton; Sailing; Biathlon; Gymnastics; Art Competitions; Alpine Skiing; Handball; Weightlifting; Wrestling; Luge; Water Polo; Hockey; Rowing; Bobsleigh; Fencing; Equestrianism; Shooting; Boxing; Taekwondo; Cycling; Diving; Canoeing; Tennis; Modern Pentathlon; Figure Skating; Golf; Softball; Archery; Volleyball; Synchronized Swimming; Table Tennis; Nordic Combined; Baseball; Rhythmic Gymnastics; Freestyle Skiing; Rugby Sevens; Trampolining; Beach Volleyball; Triathlon; Ski Jumping; Curling; Snowboarding; Rugby; Short Track Speed Skating; Skeleton; Lacrosse; Polo; Cricket; Racquets; Motorboating; Military Ski Patrol; Croquet; Jeu De Paume; Roque; Alpinism; Basque Pelota.",1180 "confidence": 4.0,1181 "notebook": "olympic-data-analysis.ipynb",1182 "id": 808,1183 "figure": null,1184 "dataset_size_mb": 8.969584465026855,1185 "dataset": "bhanupratapbiswas/olympic-data"1186 },1187 {1188 "data_type": "tabular data",1189 "domain": "Data Analysis",1190 "task_type": "Feature Engineering & Preparation, Reporting & Interpretation",1191 "language": "Python",1192 "question": "Given the file 'dataset_olympics.csv', convert the 'Year' column to a datetime type and then report the data types of all columns after this conversion.",1193 "reasoning": "First load the raw tabular dataset. Transform the temporal field ('Year') into an appropriate datetime type to enable time-aware analysis. After the conversion, inspect the schema to verify the updated data type for 'Year' and to document the types of all other columns, ensuring consistency for further processing and modeling.",1194 "answer": "Column dtypes after converting Year: ID int64; Name object; Sex object; Age object; Height object; Weight object; Team object; NOC object; Games object; Year datetime64[ns]; Season object; City object; Sport object; Event object; Medal object.",1195 "confidence": 4.0,1196 "notebook": "olympic-data-analysis.ipynb",1197 "id": 809,1198 "figure": null,1199 "dataset_size_mb": 8.969584465026855,1200 "dataset": "bhanupratapbiswas/olympic-data"