search

CNevd / libffm-ftrl

libffm with ftrl updater
BSD 3-Clause "New" or "Revised" License
93 stars 38 forks source link

how to finetune the parameters when overfitting happens #9

Closed CasyWang closed 6 years ago

CasyWang commented 7 years ago

I'm using the data from real users to make click-through rate prediction. training sample: 747265 testing sample: 261168

I got training AUC: 0.66824 testing AUC: 0.54578

It seems that the model is overfitting. how should i finetune the hyper-parameters to avoid overfitting?

CNevd commented 7 years ago

try fold cross validation