Compocyte.core.models.fit_methods

Shared fit and predict helper methods for different model wrappers.

This module centralises common training and prediction routines that are used by the different model wrappers (PyTorch, logistic regression, CatBoost etc.).

class Compocyte.core.models.fit_methods.DaskBatchDataset(*args: Any, **kwargs: Any)

Bases: IterableDataset

Compocyte.core.models.fit_methods.dataloaders_from_dask(x, y, batch_size, num_workers)

Compocyte.core.models.fit_methods.dataloaders_from_dense(x, y, batch_size, num_workers)

Compocyte.core.models.fit_methods.fit(model: DenseTorch | LogisticRegression | DummyClassifier, x: numpy.array, y: numpy.array, standardize_idx: list = None, **fit_kwargs)

Parameters:

• model (Union[DenseTorch, LogisticRegression, DummyClassifier]) – Model to be fitted.

• x (np.array) – Input data.

• y (np.array) – Target data in the shape of a 1-dimensional array of label strings.

Returns:

_description_

Return type:

_type_

Compocyte.core.models.fit_methods.fit_logreg(model: LogisticRegression, x, y, **fit_kwargs)

Compocyte.core.models.fit_methods.fit_torch(model: DenseTorch, x: numpy.array, y: numpy.array, epochs: int = 40, batch_size: int = 64, starting_lr: float = 0.01, max_lr: float = 0.1, momentum: float = 0.5, parallelize: bool = True, num_threads: int = 1, beta: float = 0.8, gamma: float = 2.0, class_balance: bool = True, max_cells: int = 1000000)

Compocyte.core.models.fit_methods.fit_trees(model: BoostedTrees, x, y, **fit_kwargs)

Compocyte.core.models.fit_methods.predict(model, x, threshold=-1, monte_carlo: int = None)

Compocyte.core.models.fit_methods.predict_logits(model, x)

Compocyte.core.models.fit_methods.samples_per_class(y)

Compocyte.core.models.fit_methods.set_threads(num_threads, parallelize)