Configuration

Training Configuration

Training behavior in DiffBench is controlled through Hydra configuration files.

For an overview of the training workflow, see Model Training.

Backend configuration

Parameters used by deep learning trainers are defined in the backend configuration group.

Typical options include:

Parameter	Description
epochs	Number of training epochs
learning_rate	Initial learning rate used by the optimizer
weight_decay	L2 regularization coefficient
debug	Enables additional debugging or logging behavior

Some models may expose additional model-specific parameters.

Model-specific configuration

Model hyperparameters are defined through the corresponding model configuration.

These can include parameters such as:

• regularization strength;

• number of estimators;

• kernel settings;

• architecture-specific options;

• optimizer settings.

For the available model families, see the Models Reference.

Hydra overrides

Configuration values can be overridden directly from the command line.

For example:

diffbenchmark-run \
    model=medicalnet \
    backend.epochs=20 \
    backend.learning_rate=1e-4

This allows experiment settings to be changed without modifying the underlying configuration files.

Analysis configuration

Analysis behavior is controlled through the analysis Hydra configuration group and command-line overrides.

Parameter	Default	Description
plots	true	Generate visualisation plots
tables	true	Build and save aggregated result tables
force_plots	false	Regenerate plots even if they already exist
analysis.debug	false	Include incomplete or failed runs when generating debug outputs

The analysis command reads experiment results from exp_outputs/experiments/ by default.

The location of experiment outputs is configured through the paths configuration.