Layer normalized with optional bias. This is based on PyTorch's :class:`~torch.nn.LayerNorm` module but is needed because PyTorch's version d

A two-layer MLP. This uses a fully-connected layer to encode the input, then applies a non-linearity, then uses another fully-connected layer

Randomly apply Gaussian blur to the image.

Randomly apply Gaussian noise to the image.

Transformation from raw image data (nanomaggies) to the rgb values displayed at the legacy viewer https://www.legacysurvey.org/viewer Co

Cross-attention head with dropout. This module is a single head of a cross-attention layer. It takes a query and a key tensor, computes the a

Augmentations tuned to the Legacy Survey Data (with minor modifications). Code copied from https://github.com/georgestein/ssl-legacysurv

MLP model

Track a series of values and provide access to smoothed values over a window or the global series average.

Normalizing flow stack for conditional distribution

Augmentations tuned to the Legacy Survey Data (with minor modifications). Code copied from https://github.com/georgestein/ssl-legacysurv

A simple feedforward neural network with 3 hidden layers.

MLP model

Modfied ResNet18.

Collection of self-attention heads. :param embedding_dim: total dimensionality of the model (equal to `head_size * num_heads`) :param

Spectrum encoder Modified version of the encoder by Serrà et al. (2018), which combines a 3 layer CNN with a dot-product attention module. Th

A transformer block, including layer norm, self-attention, another layer norm, and a two-layer MLP. :param embedding_dim: total dimensionalit

TODO: Short description of my dataset.

A cosine-annealing learning rate scheduler with initial warmup. Currently this cuts off after one cycle. The interface is otherwise compatible wi

Saves full training configuration Otherwise wandb won't log full configuration but only flattened module and data hyperparameters

PyTorch Lightning implementation of `Moco <https://arxiv.org/abs/2003.04297>`_ Paper authors: Xinlei Chen, Haoqi Fan, Ross Girshick, Kaiming He.

A dense linear layer whose parameters are tied to a tensor provided by the user. Using this layer is equivalent to using the functional form,