华为杯建模赛D赛题记录.zip

# -*- coding = utf-8 -*- # @Time :2021/10/15 12:50 下午 # @Author: XZL # @File : FS.py # @Software: PyCharm import pandas as pd from sklearn.ensemble import RandomForestRegressor from sklearn.linear_model import LinearRegression, Lasso, Ridge from sklearn.feature_selection import RFE from sklearn.svm import SVC # 删除sum为0和方差过小的数据 def delete_zero_and_var(df, var_throd): """ :param df: 输入是训练集X :param var_throd: :return: """ sum_zero_count = 0 var_count = 0 delete = [] # 删除要加上"SMILES" for index, row in df.iteritems(): if index != "SMILES" and index != "pIC50": # 计算该列和 row = (row - row.min()) / (row.max() - row.min()) if df[index].sum() == 0: df[index] = row delete.append((index)) sum_zero_count = sum_zero_count + 1 continue if row.var() < var_throd: var_count = var_count + 1 delete.append((index)) final_sum = df.shape[1] - 1 - var_count - sum_zero_count print("最终剩余特征数", final_sum) print("sum=0特征: ", sum_zero_count, "方差过小特征: ", var_count) print(delete) # 删除sum为0后的数据 handle_df = df.drop(delete, axis=1) return handle_df # 随机森林特征提取 def feature_select(df, X, y, function_name, feature_num): """ :param df: 所有总的DataFrame :param X: 需要筛选的X根据 :param y: 需要X对应的y :param feature_num: 筛选的特征数量 :return: 筛选后的DataFrame """ feature_names = list(X) # 所有列名 if function_name == "RFR": rf = RandomForestRegressor(min_samples_split=6, n_estimators=100) rf.fit(X, y) # model的feature_importances_ 参数排序 rank_impotant = sorted(zip(map(lambda x: round(x, 4), rf.feature_importances_), feature_names), reverse=True) # 展示前20个 RFR_ret = pd.DataFrame(rank_impotant).head(feature_num) return (df[list(RFR_ret[1])], list(RFR_ret[1])) if function_name == "Lasso": lasso = Lasso(alpha=.1) # alpha越大，模型越稀疏，越多特征系数变为0， lasso.fit(X, y) ret = pretty_print_linear(lasso.coef_, feature_names, sort=True) lasso_ret = pd.DataFrame(ret).head(feature_num) return df[list(lasso_ret[1])], list(lasso_ret[1]) if function_name == "RFE": lr = LinearRegression() # estimator估计函数 rfe = RFE(estimator=lr, n_features_to_select=feature_num) rfe.fit(X, y) rfe_ret = pd.DataFrame(X.columns, index=rfe.ranking_, columns=['Rank']).sort_index(ascending=True).head( feature_num) return df[list(rfe_ret['Rank'])], list(rfe_ret['Rank']) def pretty_print_linear(coefs, names=None, sort=False): if names == None: names = ["X%s" % x for x in range(len(coefs))] lst = zip(coefs, names) if sort: lst = sorted(lst, key=lambda x: x[0], reverse=True) return lst