A 17th place solution for MTS ML CUP competition using pytorch-lifestream or ptls
pip install pytorch-lifestream
- Prepare raw clicksteream for ptls format notebook
- Alternative prepare raw clicksteream for ptls format and split host to domain levels notebook
- Create some handmade features including pivots and aggregates notebook
- ALS based user factors baseline notebook
- Unsupervised pretrain COLES embeddings notebook
- Unsupervised pretrain COLES embeddings with host split notebook
- Unsupervised pretrain COLES embeddings with additional augmenation: time shuffle and dropout of sequence items notebook
- Supervised pretrain COLES embeddings notebook
- Finetune COLES embeddings to targed notebook
- Finetune COLES embeddings to targed with split hosts notebook
- Unsupervised pretrain MLM embeddings notebook
- Unsupervised pretrain tabformer embeddings notebook
- Concatenate all embeddings and features then solve downstream task notebook
- Learn separate classifiers for each features and combine to solve downstream task notebook