1、背景：http://www.datafountain.cn/#/competitions/260/intro

? ? ? ? ? ? ? ? ? DataFountain上的唯品會用戶購買行為預測比賽題目，筆者用邏輯回歸實現，分數是0.48比較弱，代碼這里參考。

2、通過比賽提取的特征如下：

? ?

特征類別	特征名	特征說明	訓練說明
基本特征	u_id	用戶唯一標識
	spu_id	商品唯一標識
	brand_id	商品所屬的品牌標識
	cat_id	商品所屬的品類標識
人的特征	u_buy_num	購買次數
	u_click_num	點擊次數
	u_buy_date	購買天數
	u_click_date	點擊天數
	u_num_ratio	購買點擊次數比，購買次數/點擊次數
	u_date_ratio	購買點擊天數比，購買天數/點擊天數
	u_buy_freq	購買頻率，購買次數/90天
	u_click_freq	點擊頻率，購買次數/90天
商品的特征	spu_buy_num	購買次數
	spu_click_num	點擊次數
	spu_buy_date	購買天數
	spu_click_date	點擊天數
	spu_num_ratio	購買點擊次數比，購買次數/點擊次數
	spu_date_ratio	購買點擊天數比，購買天數/點擊天數
	spu_buy_freq	購買頻率，購買次數/90天
	spu_click_freq	點擊頻率，購買次數/90天
品牌的特征	brand_buy_num	購買次數
	brand_click_num	點擊次數
	brand_buy_date	購買天數
	brand_click_date	點擊天數
	brand_num_ratio	購買點擊次數比，購買次數/點擊次數
	brand_date_ratio	購買點擊天數比，購買天數/點擊天數
	brand_buy_freq	購買頻率，購買次數/90天
	brand_click_freq	點擊頻率，購買次數/90天
品類的特征	cat_buy_num	購買次數
	cat_click_num	點擊次數
	cat_buy_date	購買天數
	cat_click_date	點擊天數
	cat_num_ratio	購買點擊次數比，購買次數/點擊次數
	cat_date_ratio	購買點擊天數比，購買天數/點擊天數
	cat_buy_freq	購買頻率，購買次數/90天
	cat_click_freq	點擊頻率，購買次數/90天
標記	action_type	該用戶是否會在當日購買此商品（0否，1是）

3、邏輯回歸參考代碼如下：

# -*- coding: utf-8 -*-import pandas as pd import time from sklearn import preprocessing from sklearn.linear_model import LogisticRegression from sklearn import metrics def main():#第一步：加載訓練集和測試集#加載帶標記數據label_ds=pd.read_csv(r"train_features_0714.txt",sep='\t',encoding='utf8',names=["u_id","u_buy_num","u_click_num","u_buy_date","u_click_date","u_num_ratio","u_date_ratio","u_buy_freq","u_click_freq","u_last_date",\"spu_id","spu_buy_num","spu_click_num","spu_buy_date","spu_click_date","spu_num_ratio","spu_date_ratio","spu_buy_freq","spu_click_freq","spu_last_date",\"brand_id","brand_buy_num","brand_click_num","brand_buy_date","brand_click_date","brand_num_ratio","brand_date_ratio","brand_buy_freq","brand_click_freq","brand_last_date",\"cat_id","cat_buy_num","cat_click_num","cat_buy_date","cat_click_date","cat_num_ratio","cat_date_ratio","cat_buy_freq","cat_click_freq","cat_last_date",\"action_type"]) #人特征label_ds["u_id"] = label_ds["u_id"].astype("int")label_ds["u_buy_num"] = label_ds["u_buy_num"].astype("int")label_ds["u_click_num"] = label_ds["u_click_num"].astype("int")label_ds["u_buy_date"] = label_ds["u_buy_date"].astype("int")label_ds["u_click_date"] = label_ds["u_click_date"].astype("int")label_ds["u_num_ratio"] = label_ds["u_num_ratio"].astype("float")label_ds["u_date_ratio"] = label_ds["u_date_ratio"].astype("float")label_ds["u_buy_freq"] = label_ds["u_buy_freq"].astype("float")label_ds["u_click_freq"] = label_ds["u_click_freq"].astype("float")label_ds["u_last_date"] = label_ds["u_last_date"].astype("int")#商品特征label_ds["spu_id"] = label_ds["spu_id"].astype("int")label_ds["spu_buy_num"] = label_ds["spu_buy_num"].astype("int")label_ds["spu_click_num"] = label_ds["spu_click_num"].astype("int")label_ds["spu_buy_date"] = label_ds["spu_buy_date"].astype("int")label_ds["spu_click_date"] = label_ds["spu_click_date"].astype("int")label_ds["spu_num_ratio"] = label_ds["spu_num_ratio"].astype("float")label_ds["spu_date_ratio"] = label_ds["spu_date_ratio"].astype("float")label_ds["spu_buy_freq"] = label_ds["spu_buy_freq"].astype("float")label_ds["spu_click_freq"] = label_ds["spu_click_freq"].astype("float")label_ds["spu_last_date"] = label_ds["spu_last_date"].astype("int")#品牌特征label_ds["brand_id"] = label_ds["brand_id"].astype("int")label_ds["brand_buy_num"] = label_ds["brand_buy_num"].astype("int")label_ds["brand_click_num"] = label_ds["brand_click_num"].astype("int")label_ds["brand_buy_date"] = label_ds["brand_buy_date"].astype("int")label_ds["brand_click_date"] = label_ds["brand_click_date"].astype("int")label_ds["brand_num_ratio"] = label_ds["brand_num_ratio"].astype("float")label_ds["brand_date_ratio"] = label_ds["brand_date_ratio"].astype("float")label_ds["brand_buy_freq"] = label_ds["brand_buy_freq"].astype("float")label_ds["brand_click_freq"] = label_ds["brand_click_freq"].astype("float")label_ds["brand_last_date"] = label_ds["brand_last_date"].astype("int")#品類特征label_ds["cat_id"] = label_ds["cat_id"].astype("int")label_ds["cat_buy_num"] = label_ds["cat_buy_num"].astype("int")label_ds["cat_click_num"] = label_ds["cat_click_num"].astype("int")label_ds["cat_buy_date"] = label_ds["cat_buy_date"].astype("int")label_ds["cat_click_date"] = label_ds["cat_click_date"].astype("int")label_ds["cat_num_ratio"] = label_ds["cat_num_ratio"].astype("float")label_ds["cat_date_ratio"] = label_ds["cat_date_ratio"].astype("float")label_ds["cat_buy_freq"] = label_ds["cat_buy_freq"].astype("float")label_ds["cat_click_freq"] = label_ds["cat_click_freq"].astype("float")label_ds["cat_last_date"] = label_ds["cat_last_date"].astype("int")#標記label_ds["action_type"] = label_ds["action_type"].astype("int")print "訓練集，有", label_ds.shape[0], "行", label_ds.shape[1], "列" #加載未標記數據unlabel_ds=pd.read_csv(r"test_features_0714.txt",sep='\t',encoding='utf8',names=["id","uid","spu_id","brand_id","cat_id",\"u_buy_num","u_click_num","u_buy_date","u_click_date","u_num_ratio","u_date_ratio","u_buy_freq","u_click_freq","u_last_date",\"spu_buy_num","spu_click_num","spu_buy_date","spu_click_date","spu_num_ratio","spu_date_ratio","spu_buy_freq","spu_click_freq","spu_last_date",\"brand_buy_num","brand_click_num","brand_buy_date","brand_click_date","brand_num_ratio","brand_date_ratio","brand_buy_freq","brand_click_freq","brand_last_date",\"cat_buy_num","cat_click_num","cat_buy_date","cat_click_date","cat_num_ratio","cat_date_ratio","cat_buy_freq","cat_click_freq","cat_last_date",]) #人特征unlabel_ds["id"] = unlabel_ds["id"].astype("int")unlabel_ds["u_id"] = unlabel_ds["u_id"].astype("int")unlabel_ds["u_buy_num"] = unlabel_ds["u_buy_num"].astype("int")#391萬unlabel_ds["u_click_num"] = unlabel_ds["u_click_num"].astype("int")unlabel_ds["u_buy_date"] = unlabel_ds["u_buy_date"].astype("int")unlabel_ds["u_click_date"] = unlabel_ds["u_click_date"].astype("int")unlabel_ds["u_num_ratio"] = unlabel_ds["u_num_ratio"].astype("float")unlabel_ds["u_date_ratio"] = unlabel_ds["u_date_ratio"].astype("float")unlabel_ds["u_buy_freq"] = unlabel_ds["u_buy_freq"].astype("float")unlabel_ds["u_click_freq"] = unlabel_ds["u_click_freq"].astype("float")unlabel_ds["u_last_date"] = unlabel_ds["u_last_date"].astype("int")#商品特征unlabel_ds["spu_id"] = unlabel_ds["spu_id"].astype("int")unlabel_ds["spu_buy_num"] = unlabel_ds["spu_buy_num"].astype("int")unlabel_ds["spu_click_num"] = unlabel_ds["spu_click_num"].astype("int")unlabel_ds["spu_buy_date"] = unlabel_ds["spu_buy_date"].astype("int")unlabel_ds["spu_click_date"] = unlabel_ds["spu_click_date"].astype("int")unlabel_ds["spu_num_ratio"] = unlabel_ds["spu_num_ratio"].astype("float")#241萬unlabel_ds["spu_date_ratio"] = unlabel_ds["spu_date_ratio"].astype("float")unlabel_ds["spu_buy_freq"] = unlabel_ds["spu_buy_freq"].astype("float")unlabel_ds["spu_click_freq"] = unlabel_ds["spu_click_freq"].astype("float")unlabel_ds["spu_last_date"] = unlabel_ds["spu_last_date"].astype("int")#品牌特征unlabel_ds["brand_id"] = unlabel_ds["brand_id"].astype("int")unlabel_ds["brand_buy_num"] = unlabel_ds["brand_buy_num"].astype("int")unlabel_ds["brand_click_num"] = unlabel_ds["brand_click_num"].astype("int")unlabel_ds["brand_buy_date"] = unlabel_ds["brand_buy_date"].astype("int")unlabel_ds["brand_click_date"] = unlabel_ds["brand_click_date"].astype("int")unlabel_ds["brand_num_ratio"] = unlabel_ds["brand_num_ratio"].astype("float")unlabel_ds["brand_date_ratio"] = unlabel_ds["brand_date_ratio"].astype("float")unlabel_ds["brand_buy_freq"] = unlabel_ds["brand_buy_freq"].astype("float")unlabel_ds["brand_click_freq"] = unlabel_ds["brand_click_freq"].astype("float")unlabel_ds["brand_last_date"] = unlabel_ds["brand_last_date"].astype("int")#品類特征unlabel_ds["cat_id"] = unlabel_ds["cat_id"].astype("int")unlabel_ds["cat_buy_num"] = unlabel_ds["cat_buy_num"].astype("int")unlabel_ds["cat_click_num"] = unlabel_ds["cat_click_num"].astype("int")unlabel_ds["cat_buy_date"] = unlabel_ds["cat_buy_date"].astype("int")unlabel_ds["cat_click_date"] = unlabel_ds["cat_click_date"].astype("int")unlabel_ds["cat_num_ratio"] = unlabel_ds["cat_num_ratio"].astype("float")unlabel_ds["cat_date_ratio"] = unlabel_ds["cat_date_ratio"].astype("float")unlabel_ds["cat_buy_freq"] = unlabel_ds["cat_buy_freq"].astype("float")unlabel_ds["cat_click_freq"] = unlabel_ds["cat_click_freq"].astype("float")unlabel_ds["cat_last_date"] = unlabel_ds["cat_last_date"].astype("int")print "測試集，有", unlabel_ds.shape[0], "行", unlabel_ds.shape[1], "列" #模型訓練ds_0=label_ds[label_ds['action_type']==0]#標記為0的樣本ds_0_train=ds_0.sample(frac=0.01)#抽0.01出來訓練ds_1=label_ds[label_ds['action_type']==1]#標記為1的樣本ds_train=ds_1.append(ds_0_train)label_X=ds_train[['u_num_ratio','spu_num_ratio','brand_num_ratio','cat_num_ratio']]label_X_scale=preprocessing.scale(label_X)#歸一化label_y = ds_train['action_type']#類別 ds=label_ds[label_ds['action_type']==0]model =LogisticRegression()#ensemble.GradientBoostingClassifier()model.fit(label_X_scale, label_y) #第五步：模型驗證和選擇test_df=ds_train.sample(frac=0.2)#抽0.2驗證test_X=test_df[['u_num_ratio','spu_num_ratio','brand_num_ratio','cat_num_ratio']]test_X_scale=preprocessing.scale(test_X)#歸一化test_y=test_df['action_type']#類別predicted = model.predict(test_X_scale) f1_score = metrics.f1_score(test_y, predicted) #模型評估 print f1_score#第六步：模型預測unlabe_X = unlabel_ds[['u_num_ratio','spu_num_ratio','brand_num_ratio','cat_num_ratio']]unlabe_X_scale=preprocessing.scale(unlabe_X)#歸一化unlabel_y=model.predict_proba(unlabe_X_scale)[:,1]#預測返回概率值，通過概率值閾值選擇正例樣本 out_y=pd.DataFrame(unlabel_y,columns=['prob']) #返回判定正例的比例 out_y["prob"]=out_y["prob"].apply(lambda x: '{0:.3f}'.format(x))out_1=out_y[out_y["prob"]>'0.5'] #看大于0.5的個數print out_1.shapeout_y['prob'].value_counts() #看值分布out_y.to_csv('fangjs/outvip.txt',index=False,header=None)#輸出預測數據 #執行 if __name__ == '__main__': start = time.clock() main()end = time.clock() print('finish all in %s' % str(end - start))

總結

以上是生活随笔為你收集整理的【Python学习系列二十四】scikit-learn库逻辑回归实现唯品会用户购买行为预测的全部內容，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。