hyrax.models

Submodules

Classes

`HSCAutoencoder`	This autoencoder is designed to work with datasets that are prepared with Hyrax's HSC Data Set class.
`HSCDCAE`	This autoencoder is designed to work with datasets that are prepared with Hyrax's HSC Data Set class.
`ImageDCAE`	This is an autoencoder with skipconnections that should work with
`HyraxAutoencoder`	This autoencoder is designed to work with a wide range of image datasets to allow testing.
`HyraxAutoencoderV2`	This is tweaked version of HyraxAutoencoder and is designed to work with a wide range of imaging datasets.
`HyraxCNN`	This CNN is designed to work with datasets that are prepared with Hyrax's HSC Data Set class.
`HyraxLoopback`	Simple model for testing which returns its own input

Functions

hyrax_model(cls)

Decorator to register a model with the model registry, and to add common interface functions

Package Contents

class HSCAutoencoder(config, shape)[source]

Bases: torch.nn.Module

This autoencoder is designed to work with datasets that are prepared with Hyrax’s HSC Data Set class.

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

encoder

decoder

config

forward(x)[source]

train_step(batch)[source]

This function contains the logic for a single training step. i.e. the contents of the inner loop of a ML training process.

Parameters:: batch (tuple) – A tuple containing the two values the loss function
Returns:: Current loss value – Dictionary containing the loss value for the current batch.
Return type:: dict

class HSCDCAE(config, shape)[source]

Bases: torch.nn.Module

This autoencoder is designed to work with datasets that are prepared with Hyrax’s HSC Data Set class.

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

encoder1

encoder2

encoder3

encoder4

pool

decoder4

decoder3

decoder2

decoder1

activation

config

forward(x)[source]

train_step(batch)[source]

This function contains the logic for a single training step. i.e. the contents of the inner loop of a ML training process.

Parameters:: batch (tuple) – A tuple containing the two values the loss function
Returns:: Current loss value – Dictionary containing the loss value for the current batch.
Return type:: dict

class ImageDCAE(config, shape)[source]

Bases: torch.nn.Module

This is an autoencoder with skipconnections that should work with arbitarily sized images with arbitrary number of channels.

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

input_shape

config

latent_dim

base_channel_size

conv_output_size

encoder1

encoder2

encoder3

encoder4

pool

latent_encoder

latent_decoder

decoder4

decoder3

decoder2

decoder1

activation

_calculate_conv_output_size()[source]: Calculate the output size after all convolutional layers for the linear bottleneck.

encode(x)[source]: Encode input to latent space with skip connections.

decode(latent, skip_connections, encoded_shape)[source]: Decode from latent space to image with skip connections.

forward(x)[source]: Forward pass - returns latent representation for anomaly detection.

reconstruct(x)[source]: Full reconstruction for evaluation and anomaly detection.

train_step(batch)[source]

This function contains the logic for a single training step.

Parameters:: batch (tuple) – A tuple containing the two values the loss function
Returns:: Current loss value – Dictionary containing the loss value for the current batch.
Return type:: dict

_optimizer()[source]: Default optimizer configuration.

static to_tensor(data_dict)[source]: Convert structured data to tensor format.

class HyraxAutoencoder(config, data_sample=None)[source]

Bases: torch.nn.Module

This autoencoder is designed to work with a wide range of image datasets to allow testing.

This example model is taken from this autoenocoder tutorial

The train function has been converted into train_step for use with pytorch-ignite.

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

config

c_hid

latent_dim

conv_end_w

conv_end_h

conv2d_multi_layer(input_size, num_applications, **kwargs) → int[source]

conv2d_output_size(input_size, kernel_size, padding=0, stride=1, dilation=1) → int[source]

_init_encoder()[source]

_eval_encoder(x)[source]

_init_decoder()[source]

_eval_decoder(x)[source]

forward(batch)[source]

train_step(batch)[source]

This function contains the logic for a single training step. i.e. the contents of the inner loop of a ML training process.

Parameters:: batch (tuple) – A tuple containing the inputs and labels for the current batch.
Returns:: Current loss value – Dictionary containing the loss value for the current batch.
Return type:: dict

static to_tensor(data_dict) → tuple[torch.Tensor][source]

This function converts structured data to the input tensor we need to run

Parameters:: data_dict (dict) – The dictionary returned from our data source

_optimizer()[source]

class HyraxAutoencoderV2(config, shape=(5, 250, 250))[source]

Bases: torch.nn.Module

This is tweaked version of HyraxAutoencoder and is designed to work with a wide range of imaging datasets.

V2 improvements: - Configurable final layer activation - Uses criterion and optimizer from config variables

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

config

c_hid

latent_dim

conv_end_w = 250

conv_end_h = 250

conv2d_multi_layer(input_size, num_applications, **kwargs) → int[source]

conv2d_output_size(input_size, kernel_size, padding=0, stride=1, dilation=1) → int[source]

_init_encoder()[source]

_eval_encoder(x)[source]

_init_decoder()[source]

_eval_decoder(x)[source]

forward(batch)[source]

train_step(batch)[source]

This function contains the logic for a single training step. i.e. the contents of the inner loop of a ML training process.

Parameters:: batch (tuple) – A tuple containing the inputs and labels for the current batch.
Returns:: Current loss value – Dictionary containing the loss value for the current batch.
Return type:: dict

static to_tensor(data_dict) → tuple[torch.Tensor][source]

This function converts structured data to the input tensor we need to run

Parameters:: data_dict (dict) – The dictionary returned from our data source

class HyraxCNN(config, shape=(3, 32, 32))[source]

Bases: torch.nn.Module

This CNN is designed to work with datasets that are prepared with Hyrax’s HSC Data Set class.

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

config

conv1

pool

conv2

fc1

fc2

fc3

conv2d_output_size(input_size, kernel_size, padding=0, stride=1, dilation=1) → int[source]

pool2d_output_size(input_size, kernel_size, stride, padding=0, dilation=1) → int[source]

forward(x)[source]

train_step(batch)[source]

This function contains the logic for a single training step. i.e. the contents of the inner loop of a ML training process.

Parameters:: batch (tuple) – A tuple containing the inputs and labels for the current batch.
Returns:: Current loss value – Dictionary containing the loss value for the current batch.
Return type:: dict

class HyraxLoopback(config, data_sample=None)[source]

Bases: torch.nn.Module

Simple model for testing which returns its own input

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

unused_module

config

load

forward(x)[source]: We simply return our input

train_step(batch)[source]: Training is a noop

hyrax_model(cls)[source]

Decorator to register a model with the model registry, and to add common interface functions

Returns:: The class with additional interface functions.
Return type:: type