dicee.trainer.torch_trainer

Classes

TorchTrainer

TorchTrainer for using single GPU or multi CPUs on a single node

Module Contents

class dicee.trainer.torch_trainer.TorchTrainer(args, callbacks)

Bases: dicee.abstracts.AbstractTrainer

TorchTrainer for using single GPU or multi CPUs on a single node

Arguments

callbacks: list of Abstract callback instances

loss_function = None

optimizer = None

model = None

train_dataloaders = None

training_step = None

process

fit(*args, train_dataloaders, **kwargs) → None

Training starts

Arguments

kwargs:Tuple

empty dictionary

Return type:

batch loss (float)

forward_backward_update(x_batch: torch.Tensor, y_batch: torch.Tensor) → torch.Tensor

Compute forward, loss, backward, and parameter update

Arguments

Return type:

batch loss (float)

extract_input_outputs_set_device(batch: list) → Tuple

Construct inputs and outputs from a batch of inputs with outputs From a batch of inputs and put

Arguments

Return type:

(tuple) mini-batch on select device