search

dsmilab / WSDM2018-KKBox-MRC

:musical_score: Predict the chances of a user listening to a song repetitively after the first observable listening event within a time window was triggered.
https://www.kaggle.com/c/kkbox-music-recommendation-challenge
GNU General Public License v3.0
7 stars 0 forks source link

Play_count won't improve LB score #1

Closed howeverforever closed 6 years ago

howeverforever commented 6 years ago

When I tried to add the feature where the song which how many users have listened to, local CV score rose up; however, LB score fell down.

I didn't figure out why this happened.

howeverforever commented 6 years ago

The reason is that I only consider the train data and add the feature to train set only. Furthermore, I missed calculating test data.

The edited idea is to calculate play_count from combination of train and test data. Then put it to the new column in test data. It follows that LB score boosted about 0.007. The submit could reference this commit.

howeverforever commented 6 years ago

This issue was moved to dsmilab/WSDM2018-KKBox-MRC#1