@buptlee 直接将收藏和购物车作为用户购买行为的依据却是最好么? 如果用前三个月做测试,后1个月做校验,按照你说的直接选有收藏和购物车我算的结果是 predict num is 2858 hit num is 185 total brand is 18537 precision is 0.0647305808258 call rate is 0.00998003992016 F1 is 0.0172937602244 F1才1.7%啊?
@buptlee 额,上面说的好像搞错了。上午改了一下,但是如果纯按是否有收藏和购物车来判的话,F1貌似还是不高啊 predict num is 491 hit num is 2 total brand is 1377 precision is 0.0040733197556 call rate is 0.00145243282498 F1 is 0.00214132762313 判断条件的代码是这样的: if int(op3[2])|int(op3[3]) |int(op2[2])|int(op2[3])|int(op1[2])|int(op1[3]): predict_temp.write(uid +"," + bid + "\n")