owenzhang / kaggle-avito Goto Github PK

View Code? Open in Web Editor NEW

138.0 138.0 71.0 252 KB

Winning solution to the Avito CTR competition

License: MIT License

R 100.00%

kaggle-avito's People

Stargazers

Watchers

Forkers

morandi91 albert1988 chenglongchen irwenqiang royshan ericxsun nkhuyu wangdongfrank runrunliuliu yjjsdu thanhleviet andreydrv wavelets jxlijunhao duthchao smartinsightsfromdata alexchao2012 zhiyenbay hengqujushi chuhoting beanbun jjdblast rockyzyl pjpan ysekky ashishlal gangyueqi chenguodan sandy4321 iwii0425 ninnazheng timwee venheads kevinwkc gerenuk kevinmtian rahasayantan kafkafield machinelearningorg longfly2 salomefu genesis1379 arcliu1982 hughdbrown zhugege215 arunrajeie zhydhkcws zsommer babylls ryuhou xiaoanshi vskanukolanu xinzheli628 digapieceofday khaledto huyuxiang tiffen jainnitk se4u arunkumarramanan gridl mldlx upalchowdhury hakmimi datasciences wh-forker

kaggle-avito's Issues

about the down sample.

thank you for the solution!
I have a problem about the negative down sampling.
you said :"to speed up training time, I down sample non-events at 50:1 ratio." @owenzhang
because the evaluation is logloss, not the F1 or AUC. I think the negative down sampling can make the ratio of positive and negative become very different between train and test，this can make the logloss in test become larger.

Recommend Projects