BaseModel Class

Abstract base class for Torch models in R. This class handles dataset schema parsing, output size calculation, loss function selection, and probability conversion for evaluation.

Determines output size based on label processor and task mode. Only supports single label key for now.

Selects appropriate loss function based on task type in output schema.

Converts logits into predicted probabilities for evaluation. Format depends on task mode (sigmoid or softmax, or raw). This method takes logits as input, which is a torch tensor with raw model outputs.

Usage

BaseModel(dataset)

Arguments

dataset

A dataset object (must have input_schema, output_schema, output_processors).

Value

Integer scalar representing the output dimension.

A function such as nnf_binary_cross_entropy_with_logits or nnf_cross_entropy.

Torch tensor of probabilities.

Fields

dataset

A dataset object containing input/output schema and processors.

feature_keys

Character vector of input feature names.

label_keys

Character vector of output label names.

device

Device (cpu or cuda) where the model is located.