clinicadl.train.ComputationalConfig

class clinicadl.train.ComputationalConfig(**data: Any) → None[source]

Configuration class to define computational parameters.

Parameters:

  • gpu (bool, default=False) – Whether to use a GPU.

  • non_blocking (bool, default=True) – Behavior to adopt when sending data or the model to a GPU: “When non_blocking is set to True, […] attempts to perform the conversion asynchronously with respect to the host, if possible. This asynchronous behavior applies to both pinned and pageable memory.” (see PyTorch documentation)

  • amp (bool, default=True) – Whether to use Automatic Mixed Precision.

  • channels_last (bool, default=True) – Whether to use Channels Last Memory Format when possible.

  • seed (Optional[NonNegativeInt], default=None) – Global seed to control the randomness. If None, ComputationalConfig will look for a global seed set with clinicadl.utils.seed.seed_everything(). If a seed is passed here, it will override any global seed.

  • deterministic (Optional[bool], default=None) – Whether to configure PyTorch’s operations in deterministic mode. If None, ComputationalConfig will look for a global configuration set with clinicadl.utils.seed.seed_everything(). If passed here, it will override any global configuration.