org.deeplearning4j.plot

Class NeuralNetPlotter

java.lang.Object

org.deeplearning4j.plot.NeuralNetPlotter

All Implemented Interfaces:

Serializable

public class NeuralNetPlotter
extends Object
implements Serializable

Credit to : http://yosinski.com/media/papers/Yosinski2012VisuallyDebuggingRestrictedBoltzmannMachine.pdf for visualizations

Author:

Adam Gibson

See Also:

Serialized Form

Constructor Summary

Constructors

Constructor and Description
NeuralNetPlotter()

Method Summary

Modifier and Type	Method and Description
String	getLayerGraphFilePath()
void	graphPlotType(String plotType, List<String> titles, org.nd4j.linalg.api.ndarray.INDArray[] matrices, String saveFilePath) graphPlotType sets up data to pass to scripts that render graphs
void	plotActivations(Layer layer) plotActivations show how hidden neurons are used, how often on vs.
void	plotNetworkGradient(Layer layer, Gradient gradient) plotNetworkGradient used for debugging RBM gradients with different data visualizations
void	plotNetworkGradient(Layer layer, org.nd4j.linalg.api.ndarray.INDArray gradient)
void	plotWeightHistograms(Layer network)
void	plotWeightHistograms(Layer network, Gradient gradient) plotWeightHistograms graphs values of vBias, W, and hBias on aggregate and most recent mini-batch updates (-gradient)
static void	printDataFilePath()
static void	printGraphFilePath()
void	renderFilter(Layer layer, int patchesPerRow) renderFilter plot learned filter for each hidden neuron
void	renderGraph(String action, String dataPath, String saveFilePath) Calls out to python for rendering charts
void	renderGraph(String action, String dataPath, String saveFilePath, int feature_width, int feature_height) Calls out to python for rendering charts
void	setLayerGraphFilePath(String newPath)
protected static void	setupDirectory(String path)
void	updateGraphDirectory(Layer layer)
String	writeArray(ArrayList data)
protected String	writeMatrix(org.nd4j.linalg.api.ndarray.INDArray matrix)

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Constructor Detail

NeuralNetPlotter

public NeuralNetPlotter()

Method Detail

getLayerGraphFilePath

public String getLayerGraphFilePath()

setLayerGraphFilePath

public void setLayerGraphFilePath(String newPath)

printDataFilePath

public static void printDataFilePath()

printGraphFilePath

public static void printGraphFilePath()

setupDirectory

protected static void setupDirectory(String path)

updateGraphDirectory

public void updateGraphDirectory(Layer layer)

writeMatrix

protected String writeMatrix(org.nd4j.linalg.api.ndarray.INDArray matrix)

writeArray

public String writeArray(ArrayList data)

renderGraph

public void renderGraph(String action,
                        String dataPath,
                        String saveFilePath)

Calls out to python for rendering charts

Parameters:

action - the action to take

dataPath - the path to the data

saveFilePath - the saved file path for output of graphs

renderGraph

public void renderGraph(String action,
                        String dataPath,
                        String saveFilePath,
                        int feature_width,
                        int feature_height)

Calls out to python for rendering charts

Parameters:

action - the action to take

dataPath - the path to the data

saveFilePath - the saved file path for output of graphs

feature_width - width of feature

feature_height - height of feature

graphPlotType

public void graphPlotType(String plotType,
                          List<String> titles,
                          org.nd4j.linalg.api.ndarray.INDArray[] matrices,
                          String saveFilePath)

graphPlotType sets up data to pass to scripts that render graphs

Parameters:

plotType - sets which plot to call whether "multi" for multiple matrices histograms, "hist" for a histogram of one matrix, or "scatter" for scatter plot

titles - the titles of the plots

matrices - the matrices to plot

plotWeightHistograms

public void plotWeightHistograms(Layer network,
                                 Gradient gradient)

plotWeightHistograms graphs values of vBias, W, and hBias on aggregate and most recent mini-batch updates (-gradient)

Parameters:

network - the trained neural net model

gradient - latest updates to weights and biases

plotWeightHistograms

public void plotWeightHistograms(Layer network)

plotActivations

public void plotActivations(Layer layer)

plotActivations show how hidden neurons are used, how often on vs. off and correlation

Parameters:

layer - the trained neural net layer

renderFilter

public void renderFilter(Layer layer,
                         int patchesPerRow)

renderFilter plot learned filter for each hidden neuron

Parameters:

layer - the trained neural net layer in the model

plotNetworkGradient

public void plotNetworkGradient(Layer layer,
                                Gradient gradient)

plotNetworkGradient used for debugging RBM gradients with different data visualizations

Parameters:

layer - the neural net layer

gradient - latest updates to weights and biases

plotNetworkGradient

public void plotNetworkGradient(Layer layer,
                                org.nd4j.linalg.api.ndarray.INDArray gradient)