fishy.models.deep.vae

Variational Autoencoder (VAE) for spectral classification.

Classes

class fishy.models.deep.vae.SiameseVAE(vae_backbone: VAE)

Bases: Module

Siamese VAE architecture for instance recognition.

__init__(vae_backbone: VAE) → None

Initialize internal Module state, shared by both nn.Module and ScriptModule.

forward(x1: Tensor, x2: Tensor) → Tensor

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

forward_one(x: Tensor) → Tensor

class fishy.models.deep.vae.VAE(input_dim: int, output_dim: int, hidden_dim: int = 128, num_layers: int = 4, dropout: float = 0.2, **kwargs)

Bases: Module

Variational Autoencoder (VAE) model.

input_dim

Number of input features.

Type:

int

output_dim

Number of output classes.

Type:

int

hidden_dim

Dimension of the latent space.

Type:

int

dropout

Dropout probability.

Type:

float

__init__(input_dim: int, output_dim: int, hidden_dim: int = 128, num_layers: int = 4, dropout: float = 0.2, **kwargs) → None

Initializes the VAE model.

Parameters:

• input_dim (int) – Number of input features.

• output_dim (int) – Number of output classes.

• hidden_dim (int, optional) – Latent dimension. Defaults to 128.

• num_layers (int, optional) – Number of layers. Defaults to 4.

• dropout (float, optional) – Dropout rate. Defaults to 0.2.

encode(x: Tensor) → Tuple[Tensor, Tensor]

Encodes the input tensor into mu and logvar.

forward(x: Tensor) → Tensor | Tuple[Tensor, Tensor, Tensor, Tensor]

Forward pass. Returns reconstruction and latent params if in training, or just logits.

reparameterize(mu: Tensor, logvar: Tensor) → Tensor

Reparameterization trick.

s