cvrunner.experiment package

Submodules

cvrunner.experiment.experiment module

class cvrunner.experiment.experiment.BaseExperiment

Bases: ABC

Template for the configuration. Every experiments should be stateless.

property batch_size: int: Returns: int: batch size

abstractmethod build_criterion(*args, **kwargs)

Criterion building logic

Returns:: callable criterion
Return type:: Callable

abstractmethod build_dataloader(partition)

Dataloader building logic

Parameters:: partition (str) – should be in [‘train’, ‘val’]
Return type:: DataLoader

abstractmethod build_dataset(partition)

Dataset building logic

Parameters:: partition (str) – should be in [‘train’, ‘val’]
Return type:: Dataset

abstractmethod build_loss_function()

Loss function building logic

Returns:: callable loss function
Return type:: Any

abstractmethod build_model()

Model building logic

Return type:: torch.nn.Module

abstractmethod build_optimizer_scheduler(model)

Optimizer and Learning scheduler building logic

Parameters:: model (torch.nn.Module) – model to be trained
Return type:: Tuple[Optimizer, _LRScheduler]

abstractmethod load_checkpoint(file_path)

_summary_

Parameters:: file_path (str) – checkpoint point file
Return type:: torch.nn.Module

property num_epochs: int: Returns: int: num epochs

property random_seed: int

Random seed for experiment

Returns:: Random seed
Return type:: int

abstractmethod runner_cls()

Define type of runner to run this experiment

Returns:: runner class
Return type:: BaseRunner

abstractmethod save_checkpoint()

Checkpointing logic

Return type:: None

train_epoch_end()

Train epoch ending methods

Return type:: None

train_epoch_start()

Train epoch startiting methods

Return type:: None

train_step(model, data_batch, loss_function, optimizer, lr_scheduler, device)

Train step logic

Parameters:

model (torch.nn.Module) – model to be trained
data_batch (DataBatch) – data to train
loss_function (torch.nn.Module) – loss function
optimizer (Optimizer)
lr_scheduler (_LRScheduler)

Return type:

Dict[str, Any]

val_epoch_end()

Validation epoch ending logic

Return type:: None

val_epoch_start()

Validation epoch startining methods

Return type:: None

property val_freq: int: Returns: int: validation frequency

val_step(model, data_batch, loss_function, criterion, device)

validation step logic