trainer

trojanzoo.trainer.add_argument(parser, ClassType=Trainer)

Add trainer arguments to argument parser.

For specific arguments implementation, see Trainer.add_argument().

Parameters:

• parser (argparse.ArgumentParser) – The parser to add arguments.

• ClassType (type[Trainer]) – The trainer type. Defaults to Trainer.

Returns:

argparse._ArgumentGroup – The argument group.

trojanzoo.trainer.create(dataset_name=None, dataset=None, model=None, model_ema=False, pre_conditioner=None, tensorboard=None, ClassType=Trainer, config=config, **kwargs)

Create a trainer instance.

For arguments not included in kwargs, use the default values in config.

For trainer implementation, see Trainer.

Parameters:

• dataset_name (str) – The dataset name.

• dataset (str | trojanzoo.datasets.Dataset) – Dataset instance (required for model_ema) or dataset name (as the alias of dataset_name).

• model (trojanzoo.models.Model) – Model instance.

• model_ema (bool) – Whether to use ExponentialMovingAverage. Defaults to False.

• pre_conditioner (str) –

  Choose from

  • None

  • 'kfac': KFAC

  • 'ekfac': EKFAC

  Defaults to None.

• tensorboard (bool) – Whether to use torch.utils.tensorboard.writer.SummaryWriter. Defaults to False.

• ClassType (type[Trainer]) – The trainer class type. Defaults to Trainer.

• config (Config) – The default parameter config.

• **kwargs – The keyword arguments in keys of ['optim_args', 'train_args', 'writer_args'].

Returns:

Trainer – The trainer instance.

class trojanzoo.trainer.Trainer(optim_args={}, train_args={}, writer_args={}, optimizer=None, lr_scheduler=None, model_ema=None, pre_conditioner=None, writer=None, **kwargs)

A dict-like class to contain training arguments which supports attribute-like view as well.

It inherits trojanzoo.utils.module.BasicObject.

Note

The most common usage is train(**trainer). See keys() for details.

Variables:

• optim_args (dict[str, Any]) – optimizer arguments.

• train_args (dict[str, Any]) – train function arguments.

• writer_args (dict[str, Any]) – torch.utils.tensorboard.writer.SummaryWriter arguments.

• optimizer (torch.optim.Optimizer) – Optimizer instance.

• lr_scheduler (torch.optim.lr_scheduler._LRScheduler | None) – LR_Scheduler instance.

• model_ema (ExponentialMovingAverage | None) – Exponential Moving Average instance.

• pre_conditioner (KFAC | EKFAC | None) – Pre-conditioner instance.

• writer (SummaryWriter | None) – Tensorboard summary writer instance.

classmethod add_argument(group)

Add trainer arguments to argument parser group. View source to see specific arguments.

Note

This is the implementation of adding arguments. For users, please use add_argument() instead, which is more user-friendly.

keys()

Keys include:

• All attributes exclude

  ['optim_args', 'train_args', 'writer_args']

• train_args

Returns:

list[str] – The list of keys.