Module  Description 

dopt.core.ops.basic  Contains functions for creating variable nodes and subsequently manipulating their shapes.

dopt.core.ops.math  Contains common maths operations.

dopt.core.ops.nnet  Contains common neural network operations.

dopt.core.ops.random  Contains functions for generating random numbers.

dopt.core.grads  Contains the automatic differentiation framework.

dopt.core.ops  This package facilitates the construction of various nodes in the operation graph.

dopt.nnet.layers.batchnorm  Contains an implementation of batch normalisation.

dopt.nnet.layers.conv  Contains an implementation of convolutional layers.

dopt.nnet.layers.datasource  Allows one to provide input to a network via a dopt variable.

dopt.nnet.layers.dense  Contains an implementation of dense (i.e., fully connected) layers.

dopt.nnet.layers.dropout  Contains an implementation of dropout.

dopt.nnet.layers.maxpool  Contains an implementation of max pooling.

dopt.nnet.layers.relu  Contains an implementation of the ReLU activation function.

dopt.nnet.layers.softmax  Contains an implementation of the softmat activation function.

dopt.nnet.layers  Contains generic utilities for working with Layer objects.

dopt.nnet.losses  Contains some utilities for constructing graphs for common loss functions.

dopt.nnet.networks  Provides a useful tools for constructing neural networks.

dopt.nnet.parameters  This module contains methods for initialising the parameters of neural networks.

dopt.online.adam  Contains an implementation of ADAM that relies on automatic differentiation

dopt.online.amsgrad  Contains an implementation of AMSGrad that relies on automatic differentiation

dopt.online.sgd  Contains an implementation of stochastic gradient descent that relies on automatic differentiation

dopt.core  This package contains the framework for constructing and executing operation graphs.

dopt.cpu  This module enables operation graphs to be evaluated using CPU kernels.

dopt.cuda  This is the main interface for the dopt CUDA backend.

dopt.nnet  This package contains a deep learning API backed by dopt.

dopt.online  This package contains implementations of common online optimisation algorithms, with a particular bias towards
those commonly used in large scale machine learning/deep learning.
