Simple and tsfresh features matching ones extracted in the Kaggle score=1.2

kowaalczyk / astronomical-classification

201st place (top 19%) submission for PLAsTiCC Astronomical Classification Kaggle Competition, team MIMUW E L I T E.

https://www.kaggle.com/c/PLAsTiCC-2018

4 stars 0 forks source link

Simple and tsfresh features matching ones extracted in the Kaggle score=1.2 #39

Closed kowaalczyk closed 5 years ago

kowaalczyk commented 5 years ago

DoD:

implement tsfresh and simple feature extraction in appropriate files
create script for linking these features together
parallelize

kowaalczyk commented 5 years ago

Implemented on master branch: commit 2fec200859ace96d9554ace8fdde2bf07585c64c and previous ones. Tsfresh is running now, simple aggregations are scheduled to be run later. In order to achieve model performance similar to Kaggle:

finish implementing weighted logloss evaluation metric to allow perfect early stopping
drop features that are irrelevant from joined dataset (research what is relevant and what isn't)
calculate cross-passband aggregate simple features