src.core.utils module

src.core.utils.build_model(theta: list[Tensor], model, input_dim: int) → Module

Build a model instance from the provided parameters.

Parameters:

• theta (list[torch.Tensor]) – List of model parameters.

• model_cls (Callable[[int], nn.Module]) – Model class to be instantiated.

• input_dim (int) – Input dimension of the model.

Returns:

Model instance with the provided parameters.

Return type:

nn.Module

src.core.utils.evaluate_model(name: str, model: Module, X_eval: ndarray, y_eval: ndarray) → float

Evaluate the model on the provided evaluation dataset.

Parameters:

• model (nn.Module) – Model instance to be evaluated.

• X_eval (np.ndarray) – Evaluation dataset features.

• y_eval (np.ndarray) – Evaluation dataset labels.

Returns:

Accuracy of the model on the evaluation dataset.

Return type:

float

Raises:

ValueError – If the model is not in evaluation mode.

src.core.utils.exp_delay(num_workers: int, scale: float = 0.0001) → None

src.core.utils.l1_norm(w: ndarray) → float

src.core.utils.l2_norm(w: ndarray) → float

src.core.utils.parse_args()

Parse command line arguments. User will have to choose the amount of overparametrization between 110%, 150% and 200%. :param overparam: Percentage of features vs samples. :return: Parsed arguments.

src.core.utils.set_seed(seed: int)

src.core.utils.sgd_training(X_train, y_train, num_epochs=10000, criterion=MSELoss(), batch_size=10, lr=0.01, tol=1e-08)

src.core.utils.sparsity_ratio(w: ndarray) → float

L1/L2 ratio: higher → more diffuse weights, lower → more concentrated.

src.core.utils.weight_kurtosis(w)