基于使用RFM+R模型量化用户价值进行金融产品精准营销.zip

import pandas as pd; import numpy as np import matplotlib as mpl import matplotlib.pyplot as plt mpl.rcParams["axes.unicode_minus"] = False mpl.style.use('ggplot') from sklearn import metrics from sklearn.metrics import roc_curve, auc, make_scorer, f1_score, fbeta_score, precision_score, \ roc_auc_score, accuracy_score, precision_recall_curve from sklearn.utils.multiclass import type_of_target from tools import * #################################################################################################### #################################################################################################### ###################################### ###################################### ###################################### 13. MODEL EVALUATION ###################################### ###################################### ###################################### #################################################################################################### #################################################################################################### def get_mae(depvar_value, pred_label): """ Calculate mae for predicted dataset. : params depvar_value: real label : params pred_label: predicted label """ mae = metrics.mean_absolute_error(depvar_value,pred_label) return mae def get_mse(depvar_value, pred_label): """ Calculate mse for predicted dataset. : params depvar_value: real label : params pred_label: predicted label """ mse = metrics.mean_squared_error(depvar_value,pred_label) return mse def get_rmse(depvar_value, pred_label): """ Calculate rmse for predicted dataset. : params depvar_value: real label : params pred_label: predicted label """ rmse = np.sqrt(metrics.mean_squared_error(depvar_value,pred_label)) return rmse def get_msle(depvar_value, pred_label): """ Calculate msle for predicted dataset. : params depvar_value: real label : params pred_label: predicted label """ rmse = metrics.mean_squared_log_error(depvar_value,pred_label) return rmse def get_precision(depvar_value, pred_label, pos_label=1): """ Calculate precision rate for predicted dataset. : params pred_label: real label : params pred_label: predicted label : params pos_label: positive label(0 or 1) """ precision = metrics.precision_score(depvar_value, pred_label, pos_label) return precision def get_false_neg_rate(depvar_value, pred_label, pos_label=1): """ Calculate false negative rate (eg missed fraudulent cases) for predicted dataset. : params depvar_value: real label : params pred_label: predicted label : params pos_label: positive label(0 or 1) """ tn, fp, fn, tp = metrics.confusion_matrix(depvar_value, pred_label).ravel() miss_neg_rate = fn / (fn + tn) # 误报为负占所有预测为负的比例 return miss_neg_rate def get_auc_ks(depvar_value, pred_value, pos_label=1): """ Calculate AUC and KS for predicted dataset. : params depvar_value: real label : params pred_value: predicted value : params pos_label: positive label(0 or 1) """ fpr, tpr, thr = metrics.roc_curve(depvar_value, pred_value, pos_label=pos_label) ks_list = tpr - fpr indx = np.argmax(ks_list) thr_point = thr[indx] if pos_label == 0: pred_value = 1 - pred_value pred_label = pd.Series(pred_value).map(lambda x: 1 if x > thr_point else 0) ac = metrics.auc(fpr, tpr) ks = max(tpr - fpr) return pred_label, ac, ks, thr_point, fpr, tpr def get_pos_neg_rate(pred_label): """ Calculate positive rate and negative rate for predicted dataset. : params pred_label: predicted label """ neg_rate = sum(pred_label == 0) / len(pred_label) pos_rate = sum(pred_label == 1) / len(pred_label) return neg_rate, pos_rate def get_f_beta(depvar_value, pred_label, pos_label, beta): """ Calculate positive rate and negative rate for predicted dataset. : params depvar_value: real label : params pred_label: predicted label : params pos_label: positive label(0 or 1) : params beta: beta for calculate f_beta """ f_beta = metrics.fbeta_score(depvar_value, pred_label, beta=beta) return f_beta def get_clf_statistics(pred_value, depvar_value, pos_label=1, beta=1, extra_stat=False, silent=True): """ Calculate all statistics for predicted value if label is binary. : params depvar_value: real label : params pred_value: predicted value : params pos_label: positive label(0 or 1) : params beta: beta for calculate f_beta : params extra_stat: determine if extra statistics is needed : params silent: If True, restrict prints """ try: pred_label, ac, ks, thr_point, fpr, tpr = get_auc_ks(depvar_value, pred_value, pos_label) except Exception as e: pred_label, ac, ks, thr_point, fpr, tpr = None, None, None, None, None, None print(e) try: miss_neg_rate = get_false_neg_rate(depvar_value, pred_label, pos_label) except Exception as e: miss_neg_rate = None print(e) try: neg_rate, pos_rate = get_pos_neg_rate(pred_label) except Exception as e: neg_rate, pos_rate = None, None print(e) if extra_stat: try: recall = get_recall(depvar_value, pred_label, pos_label) except Exception as e: recall = None print(e) try: precision = get_precision(depvar_value, pred_label, pos_label) except Exception as e: precision = None print(e) try: f_beta = get_f_beta(depvar_value, pred_label, pos_label, beta) except Exception as e: f_beta = None print(e) if not silent: print('outputs are auc, ks, threshold, negativerate, missing_negative_rate, recall rate, \ precision rate, f_beta, fpr, tpr') return ac, ks, thr_point, pos_rate, neg_rate, miss_neg_rate, recall, precision, f_beta, fpr, tpr else: if not silent: print('outputs are auc, ks, threshold, negativerate, missing_negative_rate, fpr, tpr') return ac, ks, thr_point, neg_rate, miss_neg_rate, fpr, tpr def get_reg_statistics(pred_value, depvar_value, silent=True): """ Calculate all statistics for predicted value if label is continuous. : params depvar_value: real label : params pred_value: predicted value : params silent: If True, restrict prints """ try: mae = get_mae(depvar_value, pred_value) except Exception as e: mae = None print(e) try: mse = get_mse(depvar_value, pred_value) except Exception as e: mse = None print(e) try: msle = get_msle(depvar_value, pred_value) except Exception as e: msle = None print(e) try: rmse = get_rmse(depvar_value, pred_value) except Exception as e: rmse = None print(e) if not silent: print('outputs are mae, mse, msle, rmse') return mae, mse, msle, rmse def get_xgb_fi(model, method='interaction', alpha=0.7, top=20): """ Get feature importance from xgb model. : params model: input model : params method: method for choosing features : params alpha: weight parameter for importance_type (larger the alpha, larger the weight for importance_type = gain) : params top: number of top features from each importance_type """ importance_by_weight = model.get_score(importance_type='weight') importance_by_gain = model.get_score(importance_type='gain') importance_by_weight = pd.DataFrame(importance_by_weight, index=range(1)).T importance_by_weight.columns = {'weight'} importance_by_weight = importance_by_weight.sort