Training

Utilities to train a model.

class ttools.training.ModelInterface

An adapter to run or train a model.

backward(batch, forward_data)

Computes gradients, take an optimizer step and update the model.

Parameters:

• batch (dict) – batch of data provided by a data pipeline.
• forward_data (dict) – outputs from the forward pass

Returns:

a dictionary of outputs

Return type:

backward_data (dict)

finalize_validation(running_data)

Computes the final validation aggregates from the running data.

The default implementation is a no-op

Parameters:running_data (dict) – current aggregates of the validation loop. Returns:initialized values Return type:validation_data (dict)

forward(batch)

Runs the model on a batch of data.

Parameters:batch (dict) – batch of data provided by a data pipeline. Returns:a dictionary of outputs Return type:forward_data (dict)

init_validation()

Initializes the quantities to be reported during validation.

The default implementation is a no-op

Returns:initialized values Return type:data (dict)

update_validation(batch, fwd, running_data)

Updates the running val data using the current batch’s forward output.

The default implementation is a no-op

Parameters:

• batch (dict) – batch of data provided by a data pipeline.
• fwd (dict) – data from one forward step in validation mode
• running_data (dict) – current aggregates of the validation loop.

Returns:

initialized values

Return type:

updated_data (dict)

class ttools.training.Trainer(interface)

Implements a simple training loop with hooks for callbacks.

Parameters:interface (ModelInterface) – adapter to run forward and backward pass on the model being trained.

callbacks

hooks that will be called while training progresses.

Type:list of Callbacks

add_callback(callback)

Adds a callback to the list of training hooks.

train(dataloader, num_epochs=None, val_dataloader=None)

Main training loop. This starts the training procedure.

Parameters:

• dataloader (DataLoader) – loader that yields training batches.
• num_epochs (int, optional) – max number of epochs to run.
• val_dataloader (DataLoader, optional) – loader that yields validation batches

class ttools.training.Checkpointer(root, model=None, meta=None, optimizers=None, prefix=None)

Save and restore model and optimizer variables.

Parameters:

• root (string) – path to the root directory where the files are stored.
• model (torch.nn.Module) –
• optimizers (list of torch.optimizer) – optimizers whose parameters will be checkpointed together with the model.
• meta (dict) –

delete(path)

Delete checkpoint at path.

load(path)

Loads a checkpoint, updates the model and returns extra data.

Parameters:path (string) – path to the checkpoint file, relative to the root dir. Returns:extra information passed by the user at save time. meta (dict): metaparameters of the model passed at save time. Return type:extras (dict)

load_latest()

Try to load the most recent checkpoint, skip failing files.

Returns:extra information passed by the user at save time. meta (dict): metaparameters of the model passed at save time. Return type:extras (dict)

static load_meta(root, prefix=None)

Fetch model metadata without touching the saved parameters.

save(path, extras=None)

Save model, metaparams and extras to relative path.

Parameters:

• path (string) – relative path to the file being saved (without extension).
• extras (dict) – extra user-provided information to be saved with the model.

sorted_checkpoints()

Get list of all checkpoints in root directory, sorted by creation date.

class ttools.training.BasicArgumentParser(*args, **kwargs)

A basic argument parser with commonly used training options.