import pandas as pd import numpy as np import streamlit as st from sklearn.impute import KNNImputer,SimpleImputer,IterativeImputer import best_tts, evaluationer,models from sklearn.experimental import enable_iterative_imputer from sklearn.model_selection import train_test_split as tts from collections import Counter from sklearn.preprocessing import PolynomialFeatures from sklearn.metrics import root_mean_squared_error import seaborn as sns from sklearn.decomposition import PCA import grid_search_cv import matplotlib.pyplot as plt import outliers,best_tts import feature_selections def Auto_optimizer(X,y,eva,model,model_name,test= None): if st.button("Train Regression Model"): num_cols = X.select_dtypes(exclude = "O").columns cat_cols = X.select_dtypes(include = "O").columns st.write("Num_cols",tuple(num_cols)) st.write("cat_cols",tuple(cat_cols)) # check for Duplicate and drop duplicated in X if len(X.isnull().sum()[(X.isnull().sum()/len(X)*100) >40]) >0: X = X.drop(columns = X.isnull().sum()[(X.isnull().sum()/len(X)*100) >40].index) st.write("Columns with more than 40% null values removed") # st.write("csx",X) len_null = X.isnull().sum().sum() st.write(f"There are {len_null} null values in Train") knn_imputed_num_X = X.copy() si_mean_imputed_num_X = X.copy() # st.write("sf",si_mean_imputed_num_X) si_median_imputed_num_X = X.copy() si_most_frequent_imputed_num_X = X.copy() iter_imputed_num_X = X.copy() knn_imputed_X_cat_dropped = knn_imputed_num_X.copy() si_mean_imputed_X_cat_dropped = si_mean_imputed_num_X.copy() si_median_imputed_X_cat_dropped = si_median_imputed_num_X.copy() si_most_frequent_imputed_X_cat_dropped = si_most_frequent_imputed_num_X.copy() iter_imputed_X_cat_dropped = iter_imputed_num_X.copy() if len_null >0: if X[num_cols].isnull().sum().sum() >0: knn_imputer = KNNImputer(n_neighbors = 5) knn_imputed_num_X[num_cols] = knn_imputer.fit_transform(knn_imputed_num_X[num_cols]) si_imputer = SimpleImputer(strategy = "mean") si_mean_imputed_num_X[num_cols] = si_imputer.fit_transform(si_mean_imputed_num_X[num_cols]) si_imputer = SimpleImputer(strategy = "median") si_median_imputed_num_X[num_cols] = si_imputer.fit_transform(si_median_imputed_num_X[num_cols]) si_imputer = SimpleImputer(strategy = "most_frequent") si_most_frequent_imputed_num_X[num_cols] = si_imputer.fit_transform(si_most_frequent_imputed_num_X[num_cols]) iter_imputer = IterativeImputer(max_iter = 200,random_state= 42) iter_imputed_num_X[num_cols] = iter_imputer.fit_transform(iter_imputed_num_X[num_cols]) knn_imputed_X_cat_dropped = knn_imputed_num_X.copy() si_mean_imputed_X_cat_dropped = si_mean_imputed_num_X.copy() si_median_imputed_X_cat_dropped = si_median_imputed_num_X.copy() si_most_frequent_imputed_X_cat_dropped = si_most_frequent_imputed_num_X.copy() iter_imputed_X_cat_dropped = iter_imputed_num_X.copy() if X[cat_cols].isnull().sum().sum() >0: # treating missing values in categorical columns # st.write("si_mean_imputed_num_X",si_mean_imputed_num_X) si_imputer = SimpleImputer(strategy = "most_frequent") knn_imputed_num_X[cat_cols] = si_imputer.fit_transform(knn_imputed_num_X[cat_cols]) si_imputer = SimpleImputer(strategy = "most_frequent") si_mean_imputed_num_X.loc[:,cat_cols] = si_imputer.fit_transform(si_mean_imputed_num_X.loc[:,cat_cols]) # st.write("si_mean_imputed_num_X",si_mean_imputed_num_X) si_median_imputed_num_X[cat_cols] = si_imputer.fit_transform(si_median_imputed_num_X[cat_cols]) si_most_frequent_imputed_num_X[cat_cols] = si_imputer.fit_transform(si_most_frequent_imputed_num_X[cat_cols]) iter_imputed_num_X[cat_cols] = si_imputer.fit_transform(iter_imputed_num_X[cat_cols]) knn_imputed_X_cat_dropped = knn_imputed_X_cat_dropped.dropna() si_mean_imputed_X_cat_dropped =si_mean_imputed_X_cat_dropped.dropna() si_median_imputed_X_cat_dropped =si_median_imputed_X_cat_dropped.dropna() si_most_frequent_imputed_X_cat_dropped =si_most_frequent_imputed_X_cat_dropped.dropna() iter_imputed_X_cat_dropped =iter_imputed_X_cat_dropped.dropna() miss_val_dropped_X = X.dropna() # list of dataframes list_X_after_missing_values= [knn_imputed_num_X, si_mean_imputed_num_X, si_median_imputed_num_X, si_most_frequent_imputed_num_X, iter_imputed_num_X, knn_imputed_X_cat_dropped, si_mean_imputed_X_cat_dropped, si_median_imputed_X_cat_dropped, si_most_frequent_imputed_X_cat_dropped, iter_imputed_X_cat_dropped, miss_val_dropped_X] list_X_after_missing_values_names= ["knn_imputed_num_X", "si_mean_imputed_num_X", "si_median_imputed_num_X", "si_most_frequent_imputed_num_X", "iter_imputed_num_X", "knn_imputed_X_cat_dropped", "si_mean_imputed_X_cat_dropped", "si_median_imputed_X_cat_dropped", "si_most_frequent_imputed_X_cat_dropped", "iter_imputed_X_cat_dropped", "miss_val_dropped_X"] # st.write("si_most_frequent_imputed_num_X",si_most_frequent_imputed_num_X,) ord_enc_cols = [] ohe_enc_cols = [] if len(cat_cols) == 0: st.write("No Categorical Columns in Train") else: st.write("Select Columns for Ordinal Encoding") for column in cat_cols: selected = st.checkbox(column) if selected: st.write(f"No. of Unique value in {column} column are", X[column].nunique()) ord_enc_cols.append(column) ohe_enc_cols = set(cat_cols) -set(ord_enc_cols) ohe_enc_cols = list(ohe_enc_cols) if len(ord_enc_cols)>0: st.write("ordinal encoded columns" ,tuple(ord_enc_cols)) if len(ohe_enc_cols)>0: st.write("one hot encoded columns" ,tuple(ohe_enc_cols)) if len(ord_enc_cols)>0: ordinal_order_vals = [] for column in ord_enc_cols: unique_vals = X.dropna()[column].unique() # st.write(f"No. of Unique value in {column} column are", len(unique_vals)) ordered_unique_vals = st.multiselect("Select values in order for Ordinal Encoding",unique_vals,unique_vals) ordinal_order_vals.append(ordered_unique_vals) st.write("order of values for Ordinal Encoding",tuple(ordinal_order_vals)) if len_null > 0: for df_name, df in enumerate(list_X_after_missing_values): # st.write(f"{list_X_after_missing_values_names[df_name]}",df) from sklearn.preprocessing import OrdinalEncoder ord = OrdinalEncoder(categories=ordinal_order_vals,handle_unknown= "use_encoded_value",unknown_value = -1 ) df[ord_enc_cols] = ord.fit_transform(df[ord_enc_cols]) # st.write(f"{list_X_after_missing_values_names[df_name]}",df) else : from sklearn.preprocessing import OrdinalEncoder ord = OrdinalEncoder(categories=ordinal_order_vals,handle_unknown= "use_encoded_value",unknown_value = -1 ) X[ord_enc_cols] = ord.fit_transform(X[ord_enc_cols]) st.write("Ordinal Encoding Completed ✅") if len(ohe_enc_cols)>0: if len_null > 0: for df_name, df in enumerate(list_X_after_missing_values): from sklearn.preprocessing import OneHotEncoder ohe = OneHotEncoder(sparse_output = False,handle_unknown = "ignore") pd.options.mode.chained_assignment = None df.loc[:, ohe.get_feature_names_out()] = ohe.fit_transform(df[ohe_enc_cols]) df.drop(columns = ohe_enc_cols,inplace = True) pd.options.mode.chained_assignment = 'warn' else: from sklearn.preprocessing import OneHotEncoder ohe = OneHotEncoder(sparse_output = False,handle_unknown = "ignore") pd.options.mode.chained_assignment = None X.loc[:, ohe.get_feature_names_out()] = ohe.fit_transform(X[ohe_enc_cols]) X.drop(columns = ohe_enc_cols,inplace = True) pd.options.mode.chained_assignment = 'warn' st.write("OneHot Encoding Completed ✅") if len(ohe_enc_cols)>0: if len_null > 0: for name,df in enumerate(list_X_after_missing_values): X_train,X_test,y_train,y_test = tts(df,y[df.index],test_size =.2 ,random_state = 42) # best_tts.best_tts(df,y,model,eva) evaluationer.evaluation(f"{list_X_after_missing_values_names[name]}",X_train,X_test,y_train,y_test,model,root_mean_squared_error,eva) else: X_train,X_test,y_train,y_test = tts(X,y[X.index],test_size =.2 ,random_state = 42) # best_tts.best_tts(X,y,model,eva) evaluationer.evaluation(f"baseline_model",X_train,X_test,y_train,y_test,model,root_mean_squared_error,eva) if len_null >0: for name,df in enumerate(list_X_after_missing_values): X_train,X_test,y_train,y_test = tts(df,y[df.index],test_size =.2 ,random_state = 42) evaluationer.evaluation(f"{list_X_after_missing_values_names[name]}",X_train,X_test,y_train,y_test,model,root_mean_squared_error,eva) if eva == "class": counter = Counter(y) total = sum(counter.values()) balance_ratio = {cls: count / total for cls, count in counter.items()} num_classes = len(balance_ratio) ideal_ratio = 1 / num_classes a = all(abs(ratio - ideal_ratio) <= 0.1 * ideal_ratio for ratio in balance_ratio.values()) if a == True: st.write("Balanced Dataset ✅") st.write("Using accuracy for Evaluation") value = "test_acc" else: st.write("Unbalanced Dataset ❌") st.write("Using F1 score for Evaluation") value = "test_f1" evaluationer.classification_evaluation_df.sort_values(by = value,inplace= True) name = str(evaluationer.classification_evaluation_df.iloc[-1,0]) st.write("df name",evaluationer.classification_evaluation_df.iloc[-1,0]) if len_null >0: b = list_X_after_missing_values_names.index(name) st.write("df",list_X_after_missing_values[b]) X = list_X_after_missing_values[b] if eva == "reg": st.write("Using R2 score for Evaluation",evaluationer.reg_evaluation_df) value = "test_r2" evaluationer.reg_evaluation_df.sort_values(by = value,inplace= True) name = str(evaluationer.reg_evaluation_df.iloc[-1,0]) if len_null >0: b = list_X_after_missing_values_names.index(name) st.write("df",list_X_after_missing_values[b]) X = list_X_after_missing_values[b] # Create a figure and axes num_plots = len(num_cols) cols = 2 # Number of columns in the subplot grid rows = (num_plots + cols - 1) // cols # Calculate the number of rows needed fig, axes = plt.subplots(rows, cols, figsize=(15, 5 * rows)) # Flatten the axes array for easy iteration, and remove any excess subplots axes = axes.flatten() for ax in axes[num_plots:]: fig.delaxes(ax) for i, col in enumerate(num_cols): sns.histplot(X[col], ax=axes[i],kde = True,color=sns.color_palette('Oranges', as_cmap=True)(0.7)) axes[i].set_title(col) # Adjust layout plt.tight_layout() # Show the plot in Streamlit st.pyplot(fig) # Create a figure and axes num_plots = len(num_cols) cols = 3 # Number of columns in the subplot grid rows = (num_plots + cols - 1) // cols # Calculate the number of rows needed fig, axes = plt.subplots(rows, cols, figsize=(15, 5 * rows)) # Flatten the axes array for easy iteration, and remove any excess subplots axes = axes.flatten() for ax in axes[num_plots:]: fig.delaxes(ax) for i, col in enumerate(num_cols): sns.boxplot(y=X[col], ax=axes[i],palette="magma") axes[i].set_title(col) # Adjust layout plt.tight_layout() # Show the plot in Streamlit st.pyplot(fig) outlier_cols = st.multiselect("De-Select columns for Detecting Outliers", num_cols,default= list(num_cols)) st.write("Checking for Outliers") outliers_df_X,outlier_indexes = outliers.detect_outliers(X,list(outlier_cols)) st.write("Outliers in Dataframe Summary",outliers_df_X) st.write("Columns for Outliers handling",tuple(outliers_df_X["columns name"])) select_outlier_cols = st.multiselect("Select columns for Outlier Handling",tuple(outliers_df_X["columns name"]),default =tuple(outliers_df_X["columns name"])) resultant,outlier_handled_df,outlier_handled_df_name= outliers.outlier_handling(X,y,model,outlier_indexes = outlier_indexes,outlier_cols = select_outlier_cols ,method = root_mean_squared_error,test_size = 0.2, random_state = 42,eva = "reg") st.write("outlier handling with methods",resultant) st.write("Best method with outlier handling",resultant.sort_values(by = "test_r2").tail(1).iloc[:,0].values[0]) try : st.write("Best X Data Index No.",outlier_handled_df_name.index(resultant.sort_values(by = "test_r2").tail(1).iloc[:,0].values[0])) st.write("Best X DataFrame after outlier handling ",outlier_handled_df[outlier_handled_df_name.index(resultant.sort_values(by = "test_r2").tail(1).iloc[:,0].values[0])]) X = outlier_handled_df[outlier_handled_df_name.index(resultant.sort_values(by = "test_r2").tail(1).iloc[:,0].values[0])] except : "evaluation of baseline model is better continuing with baseline model" X_train,X_test,y_train,y_test = tts(X,y[X.index],random_state = 42,test_size = 0.2) st.write("result_df",X) try: result_df_1 , feature_col, feature_col_name = feature_selections.feature_selection(X_train,X_test,y_train,y_test,model,alpha = 0.05) X = X.drop(columns = feature_col[feature_col_name.index(result_df_1.sort_values(by = "test_r2").tail(1).iloc[:,0].values[0])]) except: "evaluation by feature selection is not better than previous" try: result,X_train_b,X_test_b,y_train_b,y_test_b = best_tts.best_tts(X,y,model,eva) st.write("result_df",result) except: X_train,X_test,y_train,y_test = tts(X,y[X.index],test_size =0.2,random_state = 42) st.write("cheking with polynomial features") poly = PolynomialFeatures(degree=(2)) X_train_poly = poly.fit_transform(X_train) X_test_poly = poly.transform(X_test) result_df_2 = evaluationer.evaluation("polynomial features degree 2",X_train_poly,X_test_poly,y_train,y_test,model,root_mean_squared_error,eva) st.write("after polynomial features degree 2",evaluationer.reg_evaluation_df) poly1 = PolynomialFeatures(degree=(3)) X_train_poly1 = poly.fit_transform(X_train) X_test_poly1 = poly.transform(X_test) evaluationer.evaluation("polynomial features degree 3",X_train_poly1,X_test_poly1,y_train,y_test,model,root_mean_squared_error,eva) st.write("after polynomial features degree 3",evaluationer.reg_evaluation_df) pca = PCA(n_components=0.95) X_train_pca = pca.fit_transform(X_train) X_test_pca = pca.transform(X_test) evaluationer.evaluation("PCA",X_train_pca,X_test_pca,y_train,y_test,model,root_mean_squared_error,eva) st.write("After PCA",evaluationer.reg_evaluation_df) grid_search_cv.perform_grid_search(model,model_name,X_train,X_test,y_train,y_test,eva) st.write("best param",evaluationer.reg_evaluation_df) st.sidebar.button("click to clear evaluation metrics",evaluationer.reg_evaluation_df.drop(index = evaluationer.reg_evaluation_df.index))