EpicUsaMan

Generator output support

4

### Description ``` lazy_df = pl.scan_parquet('PATH_TO_SOME_FILE') lazy_df = lazy_df.select(...some long running logic...).batch(56).next(streaming=True) ``` So, basically it's something like syntax sugar for ``` lazy_df.head(skip + batch_size).collect(...params...) ``` But with preserving caches...

sink_parquet_all support

1

### Description Currently there is no possibility to write data on disk without materializing it when dealing with many lazy frames which are nested from one lazy frame The only...

There is few optimizers which can show better generalization and training time: AdEMAMix and AdEMAMixShampoo It can be included in repo or add some interface to change optimizer by yourself

Error displaying widget: model not found

1

Error displaying widget: model not found on Jupiter when GPU is enabled Without GPU widget is displaying

Cant't load model from checkpoint

1

RuntimeError: Error(s) in loading state_dict for BaseModel: Unexpected key(s) in state_dict: "epoch", "global_step", "pytorch-lightning_version", "state_dict", "loops", "callbacks", "optimizer_states", "lr_schedulers", "hparams_name", "hyper_parameters".

TabR

1

Include TabR architecture

Include GENDALF and GATE architecture

Process fails in Kaggle (P100/T4x2)

2

Problem: Process fails catboost version: 1.2.7 Operating System: ??? (Kaggle) CPU: ??? (Kaggle) GPU: P100 / T4x2 Params: {'learning_rate': 0.270640171567353, 'iterations': 1100, 'depth': 8, 'l2_leaf_reg': 0.33499139044072457, 'random_strength': 3.680321293260043, 'grow_policy': 'Depthwise',...