As per ModelGuesser.loadModelGuess(InputStream) but (optionally) allows copying to the specified temporary directory

EarlyStoppingResult: contains the results of the early stopping training, such as: - Why the training was terminated - Score vs. epoch - Epoch that the best model was found - Score of the best model - The best model (MultiLayerNetwork) itself

Interface for early stopping trainers

Interface for a layer of a neural network.

Created by Alex on 28/08/2016.

Activation Layer Used to apply activation on input and corresponding derivative on epsilon.

Output layer with different objective in co-occurrences for different objectives.

Created by davekale on 12/7/16.

Frozen layer freezes parameters of the layer it wraps, but allows the backpropagation to continue.

Output layer with different objective incooccurrences for different objectives.

Convolutional Neural Network Loss Layer.

3D convolution layer implementation.

Zero cropping layer for 1D convolutional neural networks.

Cropping layer for 3D convolutional neural networks.

3D deconvolution layer implementation.

2D Separable convolution layer implementation Separable convolutions split a regular convolution operation into two simpler operations, which are usually computationally more efficient.

Space to channels utility layer for convolutional input types.

Zero padding 3D layer for convolutional neural networks.

Subsampling 3D layer, used for downsampling a 3D convolution

2D Upsampling layer.

Embedding layer for sequences: feed-forward layer that expects fixed-length number (inputLength) of integers/indices per example as input, ranged from 0 to numClasses - 1.

Deep neural net normalization approach normalizes activations between layers "brightness normalization" Used for nets like AlexNet For a^i_{x,y} the activity of a neuron computed by applying kernel i at position (x,y) and applying ReLU nonlinearity, the response normalized activation b^i_{x,y} is given by: x^2 = (a^j_{x,y})^2 unitScale = (k + alpha * sum_{j=max(0, i - n/2)}^{max(N-1, i + n/2)} (a^j_{x,y})^2 ) y = b^i_{x,y} = x * unitScale**-beta gy = epsilon (aka deltas from previous layer) sumPart = sum(a^j_{x,y} * gb^j_{x,y}) gx = gy * unitScale**-beta - 2 * alpha * beta * sumPart/unitScale * a^i_{x,y} Reference:
http://www.cs.toronto.edu/~fritz/absps/imagenet.pdf
https://github.com/vlfeat/matconvnet/issues/10
Created by nyghtowl on 10/29/15.

Bidirectional is a "wrapper" layer: it wraps any uni-directional RNN layer to make it bidirectional.

Deprecated. 

LSTM layer implementation.

Recurrent Neural Network Loss Layer.

Simple RNN - aka "vanilla" RNN is the simplest type of recurrent neural network layer.

Deprecated. 

Event listener for each iteration

Update the gradient according to the configuration such as adagrad, momentum, and sparsity

Event listener for each iteration

This method allows to update Model used for inference in runtime, without queue reset

Create a Trainer based on the given parameters

This method is called at averagingFrequency

This method is called

Initialize the context

Initialize the context

Initialize the context

THe current model for the trainer

Update the current Model for the worker

A hook method for post update

A hook method for pre update.

A hook method for post update

A hook method for pre update.

Currently supports MultiLayerNetwork and ComputationGraph models.

Generate and write the crash dump to the crash dump root directory (by default, the working directory).

Write a model to a file

Write a model to an output stream

Returns a pretrained model for the given dataset, if available.