dopt.nnet

This package contains a deep learning API backed by dopt.

Working examples for how this package can be used are given in the examples/mnist.d and examples/cifar10.d files.

One would generally start by using UFCS to define a feed-forward network:

1 auto features = float32([128, 1, 28, 28]);
2 
3 auto layers = dataSource(features)
4              .dense(2_000)
5              .relu()
6              .dense(2_000)
7              .relu()
8              .dense(10)
9              .softmax();

The DAGNetwork class can then be used to traverse the resulting graph and aggregate parameters/loss terms:

auto network = new DAGNetwork([features], [layers]);

After this, one can define an objective function---there are a few standard loss functions implemented in dopt.nnet.losses:

auto labels = float32([128, 10]);

auto trainLoss = crossEntropy(layers.trainOutput, labels) + network.paramLoss;

where network.paramLoss is the sum of any parameter regularisation terms. The dopt.online package can be used to construct an updater:

auto updater = sgd([trainLoss], network.params, network.paramProj);

Finally, one can call this updater with some actual training data:

updater([
    features: Buffer(some_real_features),
    labels: Buffer(some_real_labels)
]);