BlendMMM/Simulator-UOPX

1import streamlit as st
2import pandas as pd
3import plotly.express as px
4import plotly.graph_objects as go
5from Eda_functions import format_numbers,line_plot,summary
6import numpy as np
7import re
8
9def sanitize_key(key, prefix=""):
10    # Use regular expressions to remove non-alphanumeric characters and spaces
11    key = re.sub(r'[^a-zA-Z0-9]', '', key)
12    return f"{prefix}{key}"
13
14
15def check_box(options, ad_stock_value,lag_value,num_columns=4, prefix=""):
16    num_rows = -(-len(options) // num_columns)  # Ceiling division to calculate rows
17
18    selected_options = []
19    adstock_info = {}  # Store adstock and lag info for each selected option
20    if ad_stock_value!=0:
21        for row in range(num_rows):
22            cols = st.columns(num_columns)
23            for col in cols:
24                if options:
25                    option = options.pop(0)
26                    key = sanitize_key(f"{option}_{row}", prefix=prefix)
27                    selected = col.checkbox(option, key=key)
28                    if selected:
29                        selected_options.append(option)
30                        
31                        # Input minimum and maximum adstock values
32                        adstock = col.slider('Select Adstock Range', 0.0, 1.0, ad_stock_value, step=0.05, format="%.2f",key= f"adstock_{key}" )
33
34                        # Input minimum and maximum lag values
35                        lag = col.slider('Select Lag Range', 0, 7, lag_value, step=1,key=f"lag_{key}" )
36
37                        # Create a dictionary to store adstock and lag info for the option
38                        option_info = {
39                            'adstock': adstock,
40                            'lag': lag}
41                        # Append the dictionary to the adstock_info list
42                        adstock_info[option]=option_info
43
44                    else:adstock_info[option]={
45                            'adstock': ad_stock_value,
46                            'lag': lag_value}
47
48        return selected_options, adstock_info   
49    else:
50        for row in range(num_rows):
51            cols = st.columns(num_columns)
52            for col in cols:
53                if options:
54                    option = options.pop(0)
55                    key = sanitize_key(f"{option}_{row}", prefix=prefix)
56                    selected = col.checkbox(option, key=key)
57                    if selected:
58                        selected_options.append(option)
59                        
60                        # Input minimum and maximum lag values
61                        lag = col.slider('Select Lag Range', 0, 7, lag_value, step=1,key=f"lag_{key}" )
62
63                        # dictionary to store adstock and lag info for the option
64                        option_info = {
65                            'lag': lag}
66                        # Append the dictionary to the adstock_info list
67                        adstock_info[option]=option_info
68
69                    else:adstock_info[option]={
70                            'lag': lag_value}
71                        
72        return selected_options, adstock_info 
73
74def apply_lag(X, features,lag_dict):
75    #lag_data=pd.DataFrame()
76    for col in features:
77        for lag in range(lag_dict[col]['lag'][0], lag_dict[col]['lag'][1] + 1):
78            if lag>0:
79                X[f'{col}_lag{lag}'] = X[col].shift(periods=lag, fill_value=0)
80    return X
81
82def apply_adstock(X, variable_name, decay):
83    values = X[variable_name].values
84    adstock = np.zeros(len(values))
85    
86    for row in range(len(values)):
87        if row == 0:
88            adstock[row] = values[row]
89        else:
90            adstock[row] = values[row] + adstock[row - 1] * decay
91    
92    return adstock
93
94def top_correlated_features(df,target,media_data):
95    corr_df=df.drop(target,axis=1)
96    #corr_df[target]=df[target]
97    #st.dataframe(corr_df)
98    for i in media_data:
99        #st.write(media_data[2])
100        #st.dataframe(corr_df.filter(like=media_data[2]))
101        d=(pd.concat([corr_df.filter(like=i),df[target]],axis=1)).corr()[target]
102        d=d.sort_values(ascending=False)
103        d=d.drop(target,axis=0)
104        corr=pd.DataFrame({'Feature_name':d.index,"Correlation":d.values})
105        corr.columns = pd.MultiIndex.from_product([[i], ['Feature_name', 'Correlation']])
106
107    return corr
108
109def top_correlated_features(df,variables,target):
110    correlation_df=pd.DataFrame()
111    for col in variables:
112        d=pd.concat([df.filter(like=col),df[target]],axis=1).corr()[target]
113        #st.dataframe(d)
114        d=d.sort_values(ascending=False).iloc[1:]
115        corr_df=pd.DataFrame({'Media_channel':d.index,'Correlation':d.values})
116        corr_df.columns=pd.MultiIndex.from_tuples([(col, 'Variable'), (col, 'Correlation')])
117        correlation_df=pd.concat([corr_df,correlation_df],axis=1)
118    return correlation_df
119
120def top_correlated_feature(df,variable,target):
121    d=pd.concat([df.filter(like=variable),df[target]],axis=1).corr()[target]
122    # st.dataframe(d)
123    d=d.sort_values(ascending=False).iloc[1:]
124    # st.dataframe(d)
125    corr_df=pd.DataFrame({'Media_channel':d.index,'Correlation':d.values})
126    corr_df['Adstock']=corr_df['Media_channel'].map(lambda x:x.split('_adst')[1] if len(x.split('_adst'))>1 else '-')
127    corr_df['Lag']=corr_df['Media_channel'].map(lambda x:x.split('_lag')[1][0] if len(x.split('_lag'))>1 else '-' )
128    corr_df.drop(['Correlation'],axis=1,inplace=True)
129    corr_df['Correlation']=np.round(d.values,2)
130    sorted_corr_df= corr_df.loc[corr_df['Correlation'].abs().sort_values(ascending=False).index]
131    #corr_df.columns=pd.MultiIndex.from_tuples([(variable, 'Variable'), (variable, 'Correlation')])
132    #correlation_df=pd.concat([corr_df,correlation_df],axis=1)
133    return sorted_corr_df