場景：特征選擇在模型訓練前是非常有意義的，實際上就是先期對特征相關性進行分析。

參考：http://blog.csdn.net/fjssharpsword/article/details/73550337

代碼：這里基于scikit-learn庫聯系了幾個特征選擇方法，實際學習任務當然需要掌握理論來應用，FeatureSelecton.py如下

# -*- coding: utf-8 -*-import pandas as pd from sklearn.feature_selection import SelectKBest ,chi2 from sklearn import preprocessing from sklearn.feature_selection import VarianceThreshold from sklearn.svm import LinearSVC from sklearn.feature_selection import SelectFromModel from sklearn.ensemble import ExtraTreesClassifier #加載數據 label_ds=pd.read_csv(r"lx_train_features_all.txt",sep='\t',encoding='utf8',names=['u_id','spu_id','brand_id','cat_id',\'u_spu_num','u_brand_num','u_cat_num','u_cat_spu','u_brand_spu',\'u_spu_freq','u_spu_date','u_spu_click_freq','u_spu_click_date',\'u_first_date','u_last_date','u_spu_ratio','u_ratio',\'action_type']) label_X = label_ds[['u_id','spu_id','brand_id','cat_id',\'u_spu_num','u_brand_num','u_cat_num','u_cat_spu','u_brand_spu',\'u_spu_freq','u_spu_date','u_spu_click_freq','u_spu_click_date',\'u_first_date','u_last_date','u_spu_ratio','u_ratio',\'action_type']] label_y = label_ds['action_type']#類別 min_max_scaler = preprocessing.MinMaxScaler()#范圍0-1縮放標準化 label_X_scaler=min_max_scaler.fit_transform(label_X)#單變量特征選擇-卡方檢驗，選擇相關性最高的前5個特征 X_chi2 = SelectKBest(chi2, k=5).fit_transform(label_X_scaler, label_y) print "訓練集，有", X_chi2.shape[0], "行", X_chi2.shape[1], "列" df_X_chi2=pd.DataFrame(X_chi2) feature_names = df_X_chi2.columns.tolist()#顯示列名 print feature_names#通過方差選擇特征。方差為0的特征會被自動移除。剩下的特征按設定的方差的閾值進行選擇。 sel = VarianceThreshold(threshold=.06)#設置方差的閾值為0.6 X_sel=sel.fit_transform(label_X_scaler)#選擇方差大于0.6的特征 df_X_sel=pd.DataFrame(X_chi2) feature_names = df_X_sel.columns.tolist()#顯示列名 print feature_names#基于L1的特征選擇 lsvc = LinearSVC(C=0.01, penalty="l1", dual=False).fit(label_X_scaler, label_y) model = SelectFromModel(lsvc, prefit=True) X_lsvc = model.transform(label_X_scaler) df_X_lsvc=pd.DataFrame(X_chi2) feature_names = df_X_lsvc.columns.tolist()#顯示列名 print feature_names#基于樹的特征選擇，并按重要性閾值選擇特征 clf = ExtraTreesClassifier()#基于樹模型進行模型選擇 clf = clf.fit(label_X_scaler, label_y) model = SelectFromModel(clf, threshold='1.00*mean',prefit=True)#選擇特征重要性為1倍均值的特征，數值越高特征越重要 X_trees = model.transform(label_X_scaler)#返回所選的特征 df_X_trees=pd.DataFrame(X_chi2) feature_names = df_X_trees.columns.tolist()#顯示列名 print feature_names

總結

以上是生活随笔為你收集整理的【Python学习系列十九】基于scikit-learn库进行特征选择的全部內容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網站內容還不錯，歡迎將生活随笔推薦給好友。