sklearn-docs/plot-k-means-digits

1import gradio as gr2import pandas as pd 3import numpy as np4from time import time5from sklearn import metrics6from sklearn.pipeline import make_pipeline7from sklearn.preprocessing import StandardScaler8from sklearn.cluster import KMeans9from sklearn.decomposition import PCA10from huggingface_hub import login11from datasets import load_dataset12import matplotlib.pyplot as plt13 14 15# https://scikit-learn.org/stable/auto_examples/cluster/plot_kmeans_digits.html#sphx-glr-auto-examples-cluster-plot-kmeans-digits-py16 17def display_plot(data, n_digits):18    reduced_data = PCA(n_components=2).fit_transform(data)19    kmeans = KMeans(init="k-means++", n_clusters=n_digits, n_init=4)20    kmeans.fit(reduced_data)21 22    # Step size of the mesh. Decrease to increase the quality of the VQ.23    h = 0.02  # point in the mesh [x_min, x_max]x[y_min, y_max].24 25    # Plot the decision boundary. For that, we will assign a color to each26    x_min, x_max = reduced_data[:, 0].min() - 1, reduced_data[:, 0].max() + 127    y_min, y_max = reduced_data[:, 1].min() - 1, reduced_data[:, 1].max() + 128    xx, yy = np.meshgrid(np.arange(x_min, x_max, h), np.arange(y_min, y_max, h))29 30    # Obtain labels for each point in mesh. Use last trained model.31    Z = kmeans.predict(np.c_[xx.ravel(), yy.ravel()])32 33    # Put the result into a color plot34    Z = Z.reshape(xx.shape)35    36    fig = plt.figure()37 38    plt.clf()39    plt.imshow(40        Z,41        interpolation="nearest",42        extent=(xx.min(), xx.max(), yy.min(), yy.max()),43        cmap=plt.cm.Paired,44        aspect="auto",45        origin="lower",46    )47 48    plt.plot(reduced_data[:, 0], reduced_data[:, 1], "k.", markersize=2)49    # Plot the centroids as a white X50    centroids = kmeans.cluster_centers_51    plt.scatter(52        centroids[:, 0],53        centroids[:, 1],54        marker="x",55        s=169,56        linewidths=3,57        color="w",58        zorder=10,59    )60    plt.title(61        "K-means clustering on the digits dataset (PCA-reduced data)\n"62        "Centroids are marked with white cross"63    )64    plt.xlim(x_min, x_max)65    plt.ylim(y_min, y_max)66    plt.xticks(())67    plt.yticks(())68    return fig69 70def bench_k_means(kmeans, name, data, labels):71    """Benchmark to evaluate the KMeans initialization methods.72 73    Parameters74    ----------75    kmeans : KMeans instance76        A :class:`~sklearn.cluster.KMeans` instance with the initialization77        already set.78    name : str79        Name given to the strategy. It will be used to show the results in a80        table.81    data : ndarray of shape (n_samples, n_features)82        The data to cluster.83    labels : ndarray of shape (n_samples,)84        The labels used to compute the clustering metrics which requires some85        supervision.86    """87    t0 = time()88    estimator = make_pipeline(StandardScaler(), kmeans).fit(data)89    fit_time = time() - t090    results = [name, fit_time, estimator[-1].inertia_]91 92    # Define the metrics which require only the true labels and estimator93    # labels94    clustering_metrics = [95        metrics.homogeneity_score,96        metrics.completeness_score,97        metrics.v_measure_score,98        metrics.adjusted_rand_score,99        metrics.adjusted_mutual_info_score,100    ]101    results += [m(labels, estimator[-1].labels_) for m in clustering_metrics]102 103    # The silhouette score requires the full dataset104    results += [105        metrics.silhouette_score(106            data,107            estimator[-1].labels_,108            metric="euclidean",109            sample_size=300,110        )111    ]112 113    return results114 115title = "A demo of K-Means clustering on the handwritten digits data"116def do_submit(kmeans_n_digit,random_n_digit, pca_n_digit):117    # Load the dataset118    dataset = load_dataset("sklearn-docs/digits", header=None)119    # convert dataset to pandas120    df = dataset['train'].to_pandas()121    data = df.iloc[:, :64]122    labels = df.iloc[:, 64]123 124    kmeans = KMeans(init="k-means++", n_clusters=int(kmeans_n_digit), n_init=4, random_state=0)125    results = bench_k_means(kmeans=kmeans, name="k-means++", data=data, labels=labels)126    127    df = pd.DataFrame(results).T128    numeric_cols = ['time','inertia','homo','compl','v-meas','ARI','AMI','silhouette']129    df.columns = ['init'] + numeric_cols130 131    kmeans = KMeans(init="random", n_clusters=int(random_n_digit), n_init=4, random_state=0)132    results = bench_k_means(kmeans=kmeans, name="random", data=data, labels=labels)133    df.loc[len(df.index)] = results134    135    pca = PCA(n_components=int(pca_n_digit)).fit(data)136    kmeans = KMeans(init=pca.components_, n_clusters=int(pca_n_digit), n_init=1)137    results = bench_k_means(kmeans=kmeans, name="PCA-based", data=data, labels=labels)138    df.loc[len(df.index)] = results139    df[df.columns[1:]] = df.iloc[:,1:].astype(float).round(3)140    141    df = df.T #Transpose for display142    df.columns = df.iloc[0,:].tolist()143    df = df.iloc[1:,:].reset_index()144    df.columns = ['metrics', 'k-means++', 'random', 'PCA-based']145    return display_plot(data, kmeans_n_digit), df146 147#Theme from - https://huggingface.co/spaces/trl-lib/stack-llama/blob/main/app.py148theme = gr.themes.Monochrome(149    primary_hue="indigo",150    secondary_hue="blue",151    neutral_hue="slate",152    radius_size=gr.themes.sizes.radius_sm,153    font=[gr.themes.GoogleFont("Open Sans"), "ui-sans-serif", "system-ui", "sans-serif"],154)155 156with gr.Blocks(title=title, theme=theme) as demo:157    gr.Markdown(f"## {title}")158    gr.Markdown("This demo is based on this [scikit-learn example](https://scikit-learn.org/stable/auto_examples/cluster/plot_kmeans_digits.html#sphx-glr-auto-examples-cluster-plot-kmeans-digits-py)")159    gr.Markdown("In this example we compare the various initialization strategies for K-means in terms of runtime and quality of the results.")160    gr.Markdown("As the ground truth is known here, we also apply different cluster quality metrics to judge the goodness of fit of the cluster labels to the ground truth.")161    gr.Markdown("Cluster quality metrics evaluated (see [Clustering performance evaluation](https://scikit-learn.org/stable/modules/clustering.html#clustering-evaluation) \162                for definitions and discussions of the metrics):")163    gr.Markdown("---")164    gr.Markdown(" We will be utilizing [digits](https://huggingface.co/datasets/sklearn-docs/digits) dataset. This dataset contains handwritten digits from 0 to 9. \165        In the context of clustering, one would like to group images such that the handwritten digits on the image are the same.")166 167 168    with gr.Row():169        with gr.Column(scale=0.5):170            kmeans_n_digit = gr.Slider(minimum=2, maximum=10, label="KMeans n_digits", info="n_digits is number of handwritten digits" , step=1, value=10)171            random_n_digit = gr.Slider(minimum=2, maximum=10, label="Random n_digits",  step=1, value=10)172            pca_n_digit = gr.Slider(minimum=2, maximum=10, label="PCA n_digits",step=1, value=10)173            174            plt_out = gr.Plot()175            176        with gr.Column(scale=0.5):177            sample_df = pd.DataFrame(np.zeros((9,4)),columns=['metrics', 'k-means++', 'random', 'PCA-based'])178        179            output = gr.Dataframe(sample_df, label="Clustering Metrics")180            181    with gr.Row():182            sub_btn = gr.Button("Submit")183            sub_btn.click(fn=do_submit, inputs=[kmeans_n_digit,random_n_digit, pca_n_digit], outputs=[plt_out, output])184 185demo.launch()