`run_model_diagnostics`

python

run_model_diagnostics(model: nn.Module, fresh_model: nn.Module, sample: ObservationSample, *, device: torch.device) -> ModelDiagnosticsReport

Measure a trained model against a fresh one of the same shape on the same observations.

Parameters:

model (Module) – The trained model.
fresh_model (Module) – A freshly initialised model of identical architecture, the control.
sample (ObservationSample) – Observations the trained policy visited.
device (device) – Device for the forward passes. Both models are moved to it.

Returns:

ModelDiagnosticsReport – The report.