AE

class behavenet.models.aes.AE(hparams)

Bases: BaseModel

Base autoencoder class.

This class can construct both linear and convolutional autoencoders. The linear autoencoder utilizes a single hidden layer, dense feedforward layers (i.e. not convolutional), and the encoding and decoding weights are tied to more closely resemble PCA/SVD. The convolutional autoencoder architecture is defined by various keys in the dict that serves as the constructor input. See the behavenet.models.ae_model_architecture_generator module to see examples for how this is done.

Methods Summary

build_model()	Construct the model using hparams.
forward(x[, dataset])	Process input data.
loss(data[, dataset, accumulate_grad, ...])	Calculate MSE loss for autoencoder.

Methods Documentation

build_model()

Construct the model using hparams.

forward(x, dataset=None, **kwargs)

Process input data.

Parameters:

• x (torch.Tensor object) – input data

• dataset (int) – used with session-specific io layers

Returns:

• y (torch.Tensor): output of shape (n_frames, n_channels, y_pix, x_pix)

• x (torch.Tensor): hidden representation of shape (n_frames, n_latents)

Return type:

tuple

loss(data, dataset=0, accumulate_grad=True, chunk_size=200)

Calculate MSE loss for autoencoder.

The batch is split into chunks if larger than a hard-coded chunk_size to keep memory requirements low; gradients are accumulated across all chunks before a gradient step is taken.

Parameters:

• data (dict) – batch of data; keys should include ‘images’ and ‘masks’, if necessary

• dataset (int, optional) – used for session-specific io layers

• accumulate_grad (bool, optional) – accumulate gradient for training step

• chunk_size (int, optional) – batch is split into chunks of this size to keep memory requirements low

Returns:

• ‘loss’ (float): mse loss

Return type:

dict