torchsig.models.spectrogram_models.efficientnet.efficientnet2d.EfficientNet2d

torchsig.models.spectrogram_models.efficientnet.efficientnet2d.EfficientNet2d(input_channels: int, n_features: int, efficientnet_version: str = 'b0', drop_path_rate: float = 0.2, drop_rate: float = 0.3)

Constructs and returns a 1d version of the EfficientNet model described in “EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks”.

Parameters:

• input_channels (int) – Number of 1d input channels; e.g., common practice is to split complex number time-series data into 2 channels, representing the real and imaginary parts respectively

• n_features (int) – Number of output features; should be the number of classes when used directly for classification

• efficientnet_version (str) – Specifies the version of efficientnet to use. See the timm efficientnet documentation for details. Examples are ‘b0’, ‘b1’, and ‘b4’

• drop_path_rate (float) – Drop path rate for training

• drop_rate (float) – Dropout rate for training