Best model: https://drive.google.com/drive/folders/1PRz0trU8bJ4gXRIhK27QhIL0oBp53A93
Best kaggle run: https://wandb.ai/hse-m4xig1/hse-m4xig1/reports/Best-kaggle-run--Vmlldzo3NDk2NDQ3
Tuning best run: https://wandb.ai/hse-m4xig1/hse-m4xig1/reports/Tuning-from-epoch-12--Vmlldzo3NTg4NDIw
SpEx+ model with ConvLSTMs blocks implemented in models/lstm_spex.py
. During short time training model showed a improvement comparing to basic SpEx+ tuned on streaming task. At this moment I'm still at the process of tuning the model. Latest stable checkpoint you can find here: https://drive.google.com/drive/folders/1PRz0trU8bJ4gXRIhK27QhIL0oBp53A93