Serialized Form

Package deepnetts.data.norm

Class deepnetts.data.norm.AbstractScaler

class AbstractScaler extends Object implements Serializable

Class deepnetts.data.norm.DecimalScaler

class DecimalScaler extends AbstractScaler implements Serializable

Serialized Fields

inputDivisor

float[] inputDivisor

outputDivisor

float[] outputDivisor

Class deepnetts.data.norm.MaxScaler

class MaxScaler extends AbstractScaler implements Serializable

Serialized Fields

maxInputs

TensorBase maxInputs

maxOutputs

TensorBase maxOutputs

Class deepnetts.data.norm.MinMaxScaler

class MinMaxScaler extends AbstractScaler implements Serializable

Serialized Fields

inDivider

TensorBase inDivider

maxInput

TensorBase maxInput

maxOutput

TensorBase maxOutput

minInput

TensorBase minInput

minOutput

TensorBase minOutput

outDivider

TensorBase outDivider

Class deepnetts.data.norm.RangeScaler

class RangeScaler extends AbstractScaler implements Serializable

Serialized Fields

max

float max

min

float min

range

float range

Class deepnetts.data.norm.Standardizer

class Standardizer extends AbstractScaler implements Serializable

Serialized Fields

mean

TensorBase mean

std

TensorBase std

Package deepnetts.net

Class deepnetts.net.ConvolutionalNetwork

class ConvolutionalNetwork extends NeuralNetwork<BackpropagationTrainer> implements Serializable

serialVersionUID:

7311052836990578126L

Class deepnetts.net.FeedForwardNetwork

class FeedForwardNetwork extends NeuralNetwork<BackpropagationTrainer> implements Serializable

serialVersionUID:

5819940381359274290L

Class deepnetts.net.NeuralNetwork

class NeuralNetwork extends Object implements Serializable

serialVersionUID:

1L

Serialization Methods

readObject

private void readObject(ObjectInputStream ois)
                 throws ClassNotFoundException,
IOException

Performs additional initialization after loading the network.

Parameters:

ois -

Throws:

ClassNotFoundException

IOException

Serialized Fields

inputLayer

InputLayer inputLayer

Input layer. This layer accepts external inputs and sends them to the next layer

inputWrapper

TensorBase inputWrapper

label

String label

Network's label

layers

List<AbstractLayer> layers

Collection of all layers in this network (including input(first), output(last) and hidden(in between)). As a minimum neural network must have an input and output layer.

lossFunction

LossFunction lossFunction

Loss function Loss function represents total network error for some data, and network learns by minimizing that error. Commonly used types of loss functions are Mean Squared Error for regression problems and and Cross Entropy for classification problems.

mode

Mode mode

normalizer

AbstractScaler normalizer

outputLabels

String[] outputLabels

Labels for network outputs (classes)

outputLayer

OutputLayer outputLayer

Output layer. This layer is the final step of processing network's input and its output is network's output.

preprocessing

Preprocessing<TensorBase> preprocessing

regularizationSum

float regularizationSum

trainer

T extends Trainer trainer

Package deepnetts.net.layers

Class deepnetts.net.layers.AbstractLayer

class AbstractLayer extends Object implements Serializable

serialVersionUID:

-3972836675081087082L

Serialized Fields

activation

ActivationFunction activation

Activation function for this layer.

activationType

ActivationType activationType

Type of activation function for this layer.

batchMode

boolean batchMode

batchSize

int batchSize

biases

Tensor1D biases

deltaBiases

Tensor1D deltaBiases

deltas

O extends TensorBase deltas

Deltas used for learning.

deltaWeights

W extends TensorBase deltaWeights

Weight changes for current and previous iteration.

depth

int depth

gradients

W extends TensorBase gradients

Gradients of a loss function calculates during a backward pass.

height

int height

inputs

I extends TensorBase inputs

Inputs to this layer. A reference to outputs in previous layer, or external input in input layer).

learningRate

float learningRate

Learning rate for this layer.

mode

Mode mode

momentum

float momentum

nextLayer

AbstractLayer nextLayer

Next layer in network.

optimizer

Optimizer optimizer

optimizerType

OptimizerType optimizerType

outputs

O extends TensorBase outputs

Layer outputs.

prevDeltaBiases

Tensor1D prevDeltaBiases

prevDeltaWeights

W extends TensorBase prevDeltaWeights

Weight changes for current and previous iteration.

prevLayer

AbstractLayer prevLayer

Previous layer in network.

randomWeightsType

RandomWeightsType randomWeightsType

regL1

float regL1

regL2

float regL2

trainable

boolean trainable

weights

W extends TensorBase weights

Input weight matrix / connectivity matrix for previous layer. Used in FullyConnected and OutputLayer. MaxPooling does not have Weights and ConvolutionalLayer has weights in filters.

width

int width

Class deepnetts.net.layers.ConvolutionalLayer

class ConvolutionalLayer extends AbstractLayer<Tensor3D,Tensor3D,Tensor4D> implements Serializable

serialVersionUID:

-3972836675081087082L

Serialization Methods

readObject

private void readObject(ObjectInputStream ois)
                 throws ClassNotFoundException,
IOException

Throws:

ClassNotFoundException

IOException

Serialized Fields

fCenterX

int fCenterX

fCenterY

int fCenterY

filterCache

float[][] filterCache

filterCacheCreated

boolean filterCacheCreated

filterDepth

int filterDepth

Filter depth, corresponds to number of channels in previous layer.

filterGroups

int filterGroups

filterHeight

int filterHeight

Filter height (rows)

filters

Tensor4D filters

Convolutional filters. Filters are stored as tensors.

filterWidth

int filterWidth

Convolutional filter width (columns)

inGroupSize

int inGroupSize

maxIdx

int[][][][] maxIdx

multithreaded

boolean multithreaded

outGroupSize

int outGroupSize

padding

int padding

Border padding filled with zeros (0, 1 or 2) Usually half of the filter size

stride

int stride

Convolution step, 1 by default. Number of steps convolutional filter is moved during convolution. Commonly used values 1, 2, rarely 3

useConvCache

boolean useConvCache

Class deepnetts.net.layers.FlattenLayer

class FlattenLayer extends AbstractLayer<Tensor1D,Tensor1D,TensorBase> implements Serializable

Class deepnetts.net.layers.FullyConnectedLayer

class FullyConnectedLayer extends AbstractLayer<TensorBase,TensorBase,Tensor2D> implements Serializable

serialVersionUID:

-8172689320815816927L

Serialized Fields

dropout

float dropout

useDropout

boolean useDropout

Class deepnetts.net.layers.InputLayer

class InputLayer extends AbstractLayer implements Serializable

serialVersionUID:

6519195289402298178L

Serialized Fields

tensorDim

int tensorDim

Class deepnetts.net.layers.MaxPoolingLayer

class MaxPoolingLayer extends AbstractLayer<Tensor3D,Tensor3D,TensorBase> implements Serializable

serialVersionUID:

6257978737942468865L

Serialized Fields

filterHeight

int filterHeight

Filter dimensions. Commonly used 2x2 with stride 2

filterWidth

int filterWidth

Filter dimensions. Commonly used 2x2 with stride 2

maxIdx

int[][][][] maxIdx

Max activation idxs. Remember idx of max output for each filter position. [channel][row][col][2]

stride

int stride

Filter step. Commonly used 2

Class deepnetts.net.layers.OutputLayer

class OutputLayer extends AbstractLayer<TensorBase,TensorBase,Tensor2D> implements Serializable

serialVersionUID:

4319240027730054207L

Serialized Fields

labels

String[] labels

lossType

LossType lossType

outputErrors

TensorBase outputErrors

singleOutInput

float singleOutInput

Class deepnetts.net.layers.SoftmaxOutputLayer

class SoftmaxOutputLayer extends OutputLayer implements Serializable

serialVersionUID:

609777460047517229L

Package deepnetts.net.layers.activation

Class deepnetts.net.layers.activation.LeakyRelu

class LeakyRelu extends Object implements Serializable

Serialized Fields

a

float a

Class deepnetts.net.layers.activation.Linear

class Linear extends Object implements Serializable

Serialized Fields

slope

int slope

Class deepnetts.net.layers.activation.Relu

class Relu extends Object implements Serializable

Class deepnetts.net.layers.activation.Sigmoid

class Sigmoid extends Object implements Serializable

Class deepnetts.net.layers.activation.SoftSign

class SoftSign extends Object implements Serializable

Class deepnetts.net.layers.activation.Tanh

class Tanh extends Object implements Serializable

Package deepnetts.net.loss

Class deepnetts.net.loss.BinaryCrossEntropyLoss

class BinaryCrossEntropyLoss extends Object implements Serializable

serialVersionUID:

3236227875327534698L

Serialized Fields

outputError

float[] outputError

Cross entropy derivative is dE/dy = (y-t) / y*(1-y) Since denominator is same as sigmoid derivative, they are canceled when calculating delta in output layer: delta = dE/dy * dy/ds = y - t See http://neuralnetworksanddeeplearning.com/chap3.html#introducing_the_cross-entropy_cost_function http://peterroelants.github.io/posts/neural_network_implementation_intermezzo01/

outputErrorTsr

TensorBase outputErrorTsr

outputLayer

OutputLayer outputLayer

patternCount

float patternCount

patternLoss

float patternLoss

regularizationSum

float regularizationSum

totalError

float totalError

Class deepnetts.net.loss.CrossEntropyLoss

class CrossEntropyLoss extends Object implements Serializable

serialVersionUID:

3646652383636911829L

Serialized Fields

outputError

float[] outputError

Sum over all outputs and training samples Bishop, pg. 245, eq. 6.185 dE/ds = (y - t) (when Softmax activation is used) http://peterroelants.github.io/posts/neural_network_implementation_intermezzo02/

outputErrorTsr

TensorBase outputErrorTsr

patternCount

int patternCount

patternLoss

float patternLoss

patternLossTsr

TensorBase patternLossTsr

regularizationSum

float regularizationSum

targetIdx

int targetIdx

totalError

float totalError

Class deepnetts.net.loss.MeanSquaredErrorLoss

class MeanSquaredErrorLoss extends Object implements Serializable

serialVersionUID:

5004545721099801809L

Serialized Fields

outputError

float[] outputError

outputErrorTsr

TensorBase outputErrorTsr

patternCount

int patternCount

patternLoss

float patternLoss

patternLossTsr

TensorBase patternLossTsr

regularizationSum

float regularizationSum

totalError

float totalError

Package deepnetts.net.train

Class deepnetts.net.train.BackpropagationTrainer

class BackpropagationTrainer extends Object implements Serializable

Serialization Methods

readObject

private void readObject(ObjectInputStream ois)
                 throws ClassNotFoundException,
IOException

This method needs to be overridden to initialize transient fields after deserialization.

Parameters:

ois -

Throws:

ClassNotFoundException

IOException

Serialized Fields

avgConvergenceSpeeed

float[] avgConvergenceSpeeed

batchMode

boolean batchMode

Set to true to use batch mode training

batchSize

int batchSize

Size of mini batch. When full batch is used, this equals training set size

checkpointEpochs

int checkpointEpochs

How many epochs for early stopping checkpoint.

dropout

float dropout

earlyStopping

boolean earlyStopping

Use early stopping setting.

earlyStoppingCheckpointCount

int earlyStoppingCheckpointCount

earlyStoppingMinLossChange

float earlyStoppingMinLossChange

Min delta between checkpoints to continue training

earlyStoppingPatience

int earlyStoppingPatience

How many checkpoints to wait before stopping training

epoch

int epoch

Current training epoch

extendedLogging

boolean extendedLogging

learningRate

float learningRate

Global learning rate, which controls the step size for the weights adjustment. Effectively a percent of error used to change weight.

lossFunction

LossFunction lossFunction

lossHistory

LinkedList<Float> lossHistory

maxEpochs

long maxEpochs

Training stops when this number of epochs is reached regardless of the total network error. One epoch represents one pass of the entire training set.

momentum

float momentum

Global momentum parameter

neuralNet

NeuralNetwork<?> neuralNet

A neural network to train.

optType

OptimizerType optType

Optimization algorithm type

prevCheckpointLoss

float prevCheckpointLoss

regL1

float regL1

regL2

float regL2

shuffle

boolean shuffle

Shuffle training set before each epoch during training.

snapshotEpochs

int snapshotEpochs

snapshotPath

String snapshotPath

stopAccuracy

float stopAccuracy

stopError

float stopError

Training stops once total training error has reached this value.

stopTraining

boolean stopTraining

Flag to stop training

totalTrainingLoss

float totalTrainingLoss

trainAccuracy

float trainAccuracy

trainingSnapshots

boolean trainingSnapshots

valAccuracy

float valAccuracy

valLoss

float valLoss

Value of loss function calculated on validation set

Package deepnetts.net.train.opt

Class deepnetts.net.train.opt.AdaGradOptimizer

class AdaGradOptimizer extends Object implements Serializable

serialVersionUID:

2833280391108408107L

Serialized Fields

learningRate

float learningRate

prevBiasSqrSum

Tensor1D prevBiasSqrSum

prevGradSqrSum

TensorBase prevGradSqrSum

prevGradSqrSum2D

Tensor2D prevGradSqrSum2D

prevGradSqrSum4D

Tensor4D prevGradSqrSum4D

Class deepnetts.net.train.opt.AdamOptimizer

class AdamOptimizer extends Object implements Serializable

serialVersionUID:

6890113339772540541L

Serialized Fields

beta1

float beta1

beta2

float beta2

biasCorrection1

float biasCorrection1

biasCorrection2

float biasCorrection2

epoch

int epoch

learningRate

float learningRate

movAvgBiasSqrSum

Tensor1D movAvgBiasSqrSum

movAvgBiasSum

Tensor1D movAvgBiasSum

movAvgGradSqrSum

TensorBase movAvgGradSqrSum

movAvgGradSqrSum2D

Tensor2D movAvgGradSqrSum2D

movAvgGradSqrSum4D

Tensor4D movAvgGradSqrSum4D

movAvgGradSum

TensorBase movAvgGradSum

movAvgGradSum2D

Tensor2D movAvgGradSum2D

movAvgGradSum4D

Tensor4D movAvgGradSum4D

Class deepnetts.net.train.opt.MomentumOptimizer

class MomentumOptimizer extends Object implements Serializable

serialVersionUID:

6936741415174730939L

Serialized Fields

learningRate

float learningRate

momentum

float momentum

prevDeltaBiases

Tensor1D prevDeltaBiases

prevDeltaWeights

TensorBase prevDeltaWeights

prevDeltaWeights2D

Tensor2D prevDeltaWeights2D

prevDeltaWeights4D

Tensor4D prevDeltaWeights4D

Class deepnetts.net.train.opt.RmsPropOptimizer

class RmsPropOptimizer extends Object implements Serializable

serialVersionUID:

2804113940324244785L

Serialized Fields

beta

float beta

learningRate

float learningRate

movAvgBiasSqrSum

Tensor1D movAvgBiasSqrSum

movAvgGradSqrSum

TensorBase movAvgGradSqrSum

movAvgGradSqrSum2D

Tensor2D movAvgGradSqrSum2D

movAvgGradSqrSum4D

Tensor4D movAvgGradSqrSum4D

Class deepnetts.net.train.opt.SgdOptimizer

class SgdOptimizer extends Object implements Serializable

serialVersionUID:

7408169634780483571L

Serialized Fields

learningRate

float learningRate

Package deepnetts.tensor

Class deepnetts.tensor.Layout

class Layout extends Object implements Serializable

Serialized Fields

cIdx

int cIdx

hIdx

int hIdx

nIdx

int nIdx

wIdx

int wIdx

Class deepnetts.tensor.Shape

class Shape extends Object implements Serializable

Serialized Fields

dimensions

int[] dimensions

MAX_DIMENSIONS

int MAX_DIMENSIONS

numDimensions

int numDimensions

Class deepnetts.tensor.Tensor1D

class Tensor1D extends TensorBase implements Serializable

Class deepnetts.tensor.Tensor2D

class Tensor2D extends TensorBase implements Serializable

Serialized Fields

cols

int cols

colsCache

float[][] colsCache

layout

Layout layout

rows

int rows

rowsCache

float[][] rowsCache

Class deepnetts.tensor.Tensor3D

class Tensor3D extends TensorBase implements Serializable

Serialized Fields

cols

int cols

depth

int depth

rows

int rows

Class deepnetts.tensor.Tensor4D

class Tensor4D extends TensorBase implements Serializable

Serialized Fields

cols

int cols

depth

int depth

fourthDim

int fourthDim

rows

int rows

Class deepnetts.tensor.TensorBase

class TensorBase extends Object implements Serializable

serialVersionUID:

9123299065043860349L

Serialized Fields

numDimensions

int numDimensions

shape

Shape shape

values

float[] values

Values stored in this tensor make it final , only input layer and tests sets values

Package deepnetts.util

Exception Class deepnetts.util.DeepNettsException

class DeepNettsException extends RuntimeException implements Serializable

Class deepnetts.util.ImagePreprocessing

class ImagePreprocessing extends Object implements Serializable

Serialized Fields

invertPixels

boolean invertPixels

isEnabled

boolean isEnabled

mean

TensorBase mean

scaleImage

boolean scaleImage

subMean

boolean subMean

Exception Class deepnetts.util.LicenseChecker.LicenceException

class LicenceException extends RuntimeException implements Serializable

Class deepnetts.util.TypedProperties

class TypedProperties extends Properties implements Serializable