kinoml.ml.lightning_modules

Training loops built with pytorch-lightning

WIP

Module Contents

class kinoml.ml.lightning_modules.RootMeanSquaredError

Bases: pytorch_lightning.metrics.MeanSquaredError

WIP

compute()

WIP

class kinoml.ml.lightning_modules.ObservationModelModule(nn_model, optimizer, loss_function, validate=True)

Bases: pytorch_lightning.LightningModule

WIP

forward(x, observation_model=_null_observation_model)

WIP

_standard_step(batch, batch_idx, **kwargs)

WIP

training_step(batch, batch_idx, **kwargs)

WIP

_common_validation_test_step(batch, batch_idx, dataloader_idx=0, metric_prefix='val', **kwargs)

WIP

_common_validation_test_epoch_end(output_results, metric_prefix='val')

WIP

_disabled_validation_step(*args, **kwargs)

WIP

_disabled_validation_epoch_end(*args, **kwargs)

WIP

test_step(*args, **kwargs)

WIP

test_epoch_end(*args, **kwargs)

WIP

configure_optimizers()

WIP

class kinoml.ml.lightning_modules.MultiDataModule(datasets: List[torch.utils.data.Dataset], observation_models: List[callable] = (_null_observation_model,), **kwargs)

Bases: pytorch_lightning.LightningDataModule

WIP

property active_dataset

WIP

property active_dataset_index

WIP

dataset_indices_by_size(reverse=False)

WIP

_build_dataloader(dataset_index=None, indices=None)

WIP

train_dataloader(dataset_index=None)

WIP

val_dataloader(dataset_index=None)

WIP

test_dataloader(dataset_index=None)

WIP

get_kfold(nfolds=5, with_validation=True, shuffle=False, **kwargs)

WIP

class kinoml.ml.lightning_modules.CrossValidateTrainer(nfolds=5, with_validation=True, shuffle=False, *args, **kwargs)

WIP

fit(model, datamodule)

WIP

_patch_paths_for_kfold(fold_trainer, fold)

WIP

test(dataset_index=0, **kwargs)

WIP

best_run()

WIP

clear()

WIP

class kinoml.ml.lightning_modules.ObservationModelDataLoader(observation_model=_null_observation_model, *args, **kwargs)

Bases: torch.utils.data.DataLoader

WIP

__iter__()

class kinoml.ml.lightning_modules._IterWithObsModel(iterator, observation_model)

WIP

__getattr__(name: str)

__next__()

__iter__()

class kinoml.ml.lightning_modules.AttrList(*args, **kwargs)

Bases: list

A subclass of list that can accept additional attributes. Should be able to be used just like a regular list.

The problem: a = [1, 2, 4, 8] a.x = “Hey!” # AttributeError: ‘list’ object has no attribute ‘x’

The solution: a = L(1, 2, 4, 8) a.x = “Hey!” print a # [1, 2, 4, 8] print a.x # “Hey!” print len(a) # 4

You can also do these: a = L( 1, 2, 4, 8 , x=”Hey!” ) # [1, 2, 4, 8] a = L( 1, 2, 4, 8 )( x=”Hey!” ) # [1, 2, 4, 8] a = L( [1, 2, 4, 8] , x=”Hey!” ) # [1, 2, 4, 8] a = L( {1, 2, 4, 8} , x=”Hey!” ) # [1, 2, 4, 8] a = L( [2 ** b for b in range(4)] , x=”Hey!” ) # [1, 2, 4, 8] a = L( (2 ** b for b in range(4)) , x=”Hey!” ) # [1, 2, 4, 8] a = L( 2 ** b for b in range(4) )( x=”Hey!” ) # [1, 2, 4, 8] a = L( 2 ) # [2]

Taken from https://code.activestate.com/recipes/579103-python-addset-attributes-to-list/

__call__(**kwargs)

class kinoml.ml.lightning_modules.KFold3Way

Bases: sklearn.model_selection.KFold

WIP

split(X, y=None, groups=None, with_validation=True)

WIP