src.solutions.bert_finetune_solution
Module Contents
Classes
Base class for any competition solution. |
Functions
|
- class src.solutions.bert_finetune_solution.BertFinetuningPredictor(model_name='microsoft/deberta-v3-large', num_classes=6, lr=2e-05, batch_size=8, num_workers=8, max_length=512, weight_decay=0.01, accelerator='gpu', max_epochs=5, accumulate_grad_batches=4, precision=16, gradient_clip_val=1000, train_size=0.8, num_cross_val_splits=5, num_frozen_layers=20)
Bases:
src.solutions.base_solution.BaseSolutionBase class for any competition solution.
- fit(X: pandas.DataFrame, y: pandas.DataFrame, **kwargs)
- predict(X: pandas.DataFrame) pandas.DataFrame
- save(directory: Union[str, pathlib.Path]) None
Stores model to the directory.
The directory must be empty.
- load(directory: Union[str, pathlib.Path]) None
Loads model from the directory.
Initializes the solution correctly even if the config dict wasn’t specified in the constructor.
The directory should not contain files other than those produced by the save method of the same class.
- src.solutions.bert_finetune_solution.main()