loss

Loss wrapper.

Classes:

Name	Description
XpdeepLoss	Xpdeep loss wrapper.

XpdeepLoss

Xpdeep loss wrapper.

loss: AbstractLossCriterion The custom loss (ApiLoss, exported as torchArtifact) or a trusted Xpdeep custom loss (exported as ModuleState).

Parameters:

Name	Type	Description	Default
loss	AbstractLossCriterion		required

Methods:

Name	Description
from_torch	Build a loss object - infer required size from the model and fitted_schema.
to_model	Convert the loss object to a database model input.

Attributes:

Name	Type	Description
loss	AbstractLossCriterion

loss: AbstractLossCriterion

from_torch(fitted_schema: FittedSchema, model: XpdeepModel, loss: nn.Module) -> Self

Build a loss object - infer required size from the model and fitted_schema.

Source code in src/xpdeep/trainer/loss.py

@classmethod
def from_torch(cls, fitted_schema: FittedSchema, model: XpdeepModel, loss: nn.Module) -> Self:
    """Build a loss object - infer required size from the model and fitted_schema."""
    output_size = model.get_output_size(fitted_schema)

    inputs = torch.randn(*(5, *fitted_schema.target_size[1:]))
    targets = torch.randn(*(5, *output_size[1:]))
    api_loss = ApiLoss.from_torch_loss(loss=loss, predictions=inputs, targets=targets)
    return cls(loss=api_loss)

to_model() -> SerializedModuleInput

Convert the loss object to a database model input.

Source code in src/xpdeep/trainer/loss.py

def to_model(self) -> SerializedModuleInput:
    """Convert the loss object to a database model input."""
    return convert_serialized_module(self.loss.to_pydantic())