from sklearn.feature_selection import mutual_info_regression from statsmodels.stats.outliers_influence import variance_inflation_factor from sklearn.linear_model import Lasso from sklearn.linear_model import LogisticRegression from sklearn.metrics import roc_curve, auc import statsmodels.api as sm import pandas as pd import numpy as np import evaluationer import streamlit as st from sklearn.feature_selection import RFE,RFECV from sklearn.linear_model import Lasso from sklearn.feature_selection import SelectKBest, chi2, mutual_info_classif import numpy as np import pandas as pd import matplotlib.pyplot as plt from sklearn.linear_model import LogisticRegression from sklearn.feature_selection import RFE, RFECV, SelectKBest, chi2, mutual_info_classif from sklearn.model_selection import StratifiedKFold from sklearn.metrics import f1_score from sklearn.metrics import root_mean_squared_error def feature_selection(X_train, X_test,y_train,y_test,model_reg,alpha = 0.05): model = sm.OLS(y_train, sm.add_constant(X_train)) model_fit = model.fit() pval_cols = model_fit.pvalues[model_fit.pvalues > 0.05].index.tolist() coef_cols = model_fit.params[abs(model_fit.params) < 0.001].index.tolist() pval_and_coef_cols = list(set(coef_cols) | set(pval_cols)) mi_scores = mutual_info_regression(X_train, y_train) mi = pd.DataFrame() mi["col_name"] = X_train.columns mi["mi_score"] = mi_scores mi_cols = mi[mi.mi_score ==0].col_name.values.tolist() corr = X_train.corr() corru= pd.DataFrame(np.triu(corr),columns = corr.columns , index = corr.index) corr_u_cols = corru[corru[(corru > 0.5 )& (corru <1)].any()].index.tolist() corrl= pd.DataFrame(np.tril(corr),columns = corr.columns , index = corr.index) corr_l_cols = corrl[corrl[(corrl > 0.5 )& (corrl <1)].any()].index.tolist() X_new_vif = sm.add_constant(X_train) # Calculating VIF vif = pd.DataFrame() vif["variables"] = X_new_vif.columns vif["VIF"] = [variance_inflation_factor(X_new_vif.values, i) for i in range(X_new_vif.shape[1])] # st.write("gdfgdsdsdfad",vif) if len(vif[vif["variables"] == "const"]) == 1: vif = vif.drop(index = (vif[vif["variables"] == "const"].index[0])) # st.write("gdfgdsad",vif) # drop const in vif cols # vif_cols = X_new_vif.drop(columns = "const") vif_cols = vif[vif.VIF >10].variables.tolist() # lasso if alpha == "best": lasso_len = [] alpha_i = [] for i in range(1,1000,5): j = i/10000 model_lasso = Lasso(alpha=j) model_lasso.fit(X_train, y_train) col_df = pd.DataFrame({ "col_name": X_train.columns, "lasso_coef": model_lasso.coef_ }) a = len(col_df[col_df.lasso_coef ==0]) lasso_len.append(a) alpha_i.append(j) for i in zip(lasso_len,alpha_i): print(i) input_alpha = float(input("enter alpha")) model_lasso = Lasso(alpha=input_alpha) model_lasso.fit(X_train, y_train) col_df = pd.DataFrame({ "col_name": X_train.columns, "lasso_coef": model_lasso.coef_ }) lasso_cols =col_df[col_df.lasso_coef ==0].col_name.tolist() else: model_lasso = Lasso(alpha=alpha) model_lasso.fit(X_train, y_train) col_df = pd.DataFrame({ "col_name": X_train.columns, "lasso_coef": model_lasso.coef_ }) lasso_cols =col_df[col_df.lasso_coef ==0].col_name.tolist() feature_cols = [pval_cols,coef_cols,pval_and_coef_cols,mi_cols,corr_u_cols,corr_l_cols,vif_cols,lasso_cols] for col in feature_cols: try: st.write(f"{col}",X_train.drop(columns = col)) except: st.write(f"error IN col") feature_cols_name = ["pval_cols","coef_cols","pval_and_coef_cols","mi_cols","corr_u_cols","corr_l_cols","vif_cols","lasso_cols"] st.write("feature_cols", vif_cols) for i,j in enumerate(feature_cols): evaluationer.evaluation(f"{feature_cols_name[i]}" ,X_train.drop(columns = j),X_test.drop(columns = j),y_train,y_test,model_reg,method = root_mean_squared_error,eva = "reg") return evaluationer.reg_evaluation_df,feature_cols,feature_cols_name def clas_feature_selection(X_train, X_test,y_train,y_test,model,n_features_to_select = None, step=1,importance_getter='auto',refcv_graph= False,C=0.05,k = 10): global rfe_cols,rfecv_cols,lasso_cols,chi2_imp_col,mi_imp_col rfe = RFE(estimator= model,n_features_to_select = n_features_to_select,importance_getter=importance_getter, step=1) rfe.fit(X_train,y_train) rfe_cols = X_train.columns[rfe.support_] cv = StratifiedKFold(5) rfecv = RFECV(estimator=model, step=1, cv=cv, scoring="f1", min_features_to_select=1, n_jobs=-1) rfecv.fit(X_train,y_train) rfecv_cols = X_train.columns[rfecv.support_] if refcv_graph == True: n_scores = len(rfecv.cv_results_["mean_test_score"]) plt.figure() plt.xlabel("Number of features selected") plt.ylabel("Mean test f1") plt.errorbar(range(min_features_to_select, n_scores + min_features_to_select), rfecv.cv_results_["mean_test_score"], yerr=rfecv.cv_results_["std_test_score"], ) plt.grid(True) plt.title("Recursive Feature Elimination \nwith correlated features") plt.show() clf = LogisticRegression(penalty = "l1", C = C, random_state = 42, solver = "liblinear") clf.fit(X_train, y_train) lasso_cols = clf.feature_names_in_[clf.coef_[0] != 0] sk = SelectKBest(chi2, k=k) X_chi2 = sk.fit_transform(X_train, y_train) chi2_imp_col = X_train.columns[sk.get_support()] sk = SelectKBest(mutual_info_classif, k=k) X_mutual = sk.fit_transform(X_train, y_train) mi_imp_col = X_train.columns[sk.get_support()] feature_cols = [rfe_cols,rfecv_cols,lasso_cols,chi2_imp_col,mi_imp_col] feature_cols_name = ["rfe_cols","rfecv_cols","lasso_cols","chi2_imp_col","mi_imp_col"] for i,j in enumerate(feature_cols): # evaluationerevaluation(f"{feature_cols_name[i]} " ,X_train[j],X_test[j],y_train,y_test,model = model,eva = "class") evaluationer.evaluation(f"{feature_cols_name[i]}" ,X_train[j],X_test[j],y_train,y_test,model,method = root_mean_squared_error,eva = "class") return evaluationer.classification_evaluation_df , feature_cols, feature_cols_name