DenseModel

class lightautoml.ml_algo.torch_based.nn_models.DenseModel(n_in, n_out=1, block_config=[2, 2], drop_rate=0.1, num_init_features=None, compression=0.5, growth_size=256, bn_factor=2, act_fun=torch.nn.ReLU, use_bn=True, **kwargs)

Bases: Module

Realisation of ‘dense’ model.

Parameters:

• n_in (int) – Input dim.

• n_out (int) – Output dim.

• block_config (List[int]) – List of number of layers within each block

• drop_rate (Union[float, List[float]]) – Dropout rate for each layer separately or altogether.

• num_init_features (Optional[int]) – If not none add fc layer before model with certain dim.

• compression (float) – portion of neuron to drop after block.

• growth_size (int) – Output dim of every layer.

• bn_factor (float) – Dim of intermediate fc is increased times bn_factor in DenseModel layer.

• act_fun (Module) – Activation function.

• use_bn (bool) – Use BatchNorm.

forward(x)

Forward-pass.

Return type:

Tensor