org.deeplearning4j.util

Class TimeSeriesUtils

java.lang.Object

org.deeplearning4j.util.TimeSeriesUtils

public class TimeSeriesUtils
extends Object

Basic time series utils

Author:

Adam Gibson

Method Summary

Modifier and Type	Method and Description
static RNNFormat	getFormatFromRnnLayer(Layer layer) Get the RNNFormat from the RNN layer, accounting for the presence of wrapper layers like Bidirectional, LastTimeStep, etc
static INDArray	movingAverage(INDArray toAvg, int n) Calculate a moving average given the length
static Pair<INDArray,int[]>	pullLastTimeSteps(INDArray pullFrom, INDArray mask) Extract out the last time steps (2d array from 3d array input) accounting for the mask layer, if present.
static Pair<INDArray,int[]>	pullLastTimeSteps(INDArray pullFrom, INDArray mask, LayerWorkspaceMgr workspaceMgr, ArrayType arrayType) Extract out the last time steps (2d array from 3d array input) accounting for the mask layer, if present.
static INDArray	reshape2dTo3d(INDArray in, int miniBatchSize)
static INDArray	reshape2dTo3d(INDArray in, long miniBatchSize, LayerWorkspaceMgr workspaceMgr, ArrayType arrayType)
static INDArray	reshape3dTo2d(INDArray in)
static INDArray	reshape3dTo2d(INDArray in, LayerWorkspaceMgr workspaceMgr, ArrayType arrayType)
static INDArray	reshapeCnnMaskToTimeSeriesMask(INDArray timeSeriesMaskAsCnnMask, int minibatchSize) Reshape CNN-style 4d mask array of shape [seqLength*minibatch,1,1,1] to time series mask [mb,seqLength] This should match the assumptions (f order, etc) in RnnOutputLayer
static INDArray	reshapePerOutputTimeSeriesMaskTo2d(INDArray perOutputTimeSeriesMask)
static INDArray	reshapePerOutputTimeSeriesMaskTo2d(INDArray perOutputTimeSeriesMask, LayerWorkspaceMgr workspaceMgr, ArrayType arrayType)
static INDArray	reshapeTimeSeriesMaskToCnn4dMask(INDArray timeSeriesMask, LayerWorkspaceMgr workspaceMgr, ArrayType arrayType) Reshape time series mask arrays.
static INDArray	reshapeTimeSeriesMaskToVector(INDArray timeSeriesMask) Reshape time series mask arrays.
static INDArray	reshapeTimeSeriesMaskToVector(INDArray timeSeriesMask, LayerWorkspaceMgr workspaceMgr, ArrayType arrayType) Reshape time series mask arrays.
static INDArray	reshapeVectorToTimeSeriesMask(INDArray timeSeriesMaskAsVector, int minibatchSize) Reshape time series mask arrays.
static INDArray	reverseTimeSeries(INDArray in) Reverse an input time series along the time dimension
static INDArray	reverseTimeSeries(INDArray in, LayerWorkspaceMgr workspaceMgr, ArrayType arrayType) Reverse an input time series along the time dimension
static INDArray	reverseTimeSeries(INDArray in, LayerWorkspaceMgr workspaceMgr, ArrayType arrayType, RNNFormat dataFormat)
static INDArray	reverseTimeSeriesMask(INDArray mask) Reverse a (per time step) time series mask, with shape [minibatch, timeSeriesLength]
static INDArray	reverseTimeSeriesMask(INDArray mask, LayerWorkspaceMgr workspaceMgr, ArrayType arrayType) Reverse a (per time step) time series mask, with shape [minibatch, timeSeriesLength]

Methods inherited from class java.lang.Object

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

Method Detail

movingAverage

public static INDArray movingAverage(INDArray toAvg,
                                     int n)

Calculate a moving average given the length

Parameters:

toAvg - the array to average

n - the length of the moving window

Returns:

the moving averages for each row

reshapeTimeSeriesMaskToVector

public static INDArray reshapeTimeSeriesMaskToVector(INDArray timeSeriesMask)

Reshape time series mask arrays. This should match the assumptions (f order, etc) in RnnOutputLayer

Parameters:

timeSeriesMask - Mask array to reshape to a column vector

Returns:

Mask array as a column vector

reshapeTimeSeriesMaskToVector

public static INDArray reshapeTimeSeriesMaskToVector(INDArray timeSeriesMask,
                                                     LayerWorkspaceMgr workspaceMgr,
                                                     ArrayType arrayType)

Reshape time series mask arrays. This should match the assumptions (f order, etc) in RnnOutputLayer

Parameters:

timeSeriesMask - Mask array to reshape to a column vector

Returns:

Mask array as a column vector

reshapeTimeSeriesMaskToCnn4dMask

public static INDArray reshapeTimeSeriesMaskToCnn4dMask(INDArray timeSeriesMask,
                                                        LayerWorkspaceMgr workspaceMgr,
                                                        ArrayType arrayType)

Reshape time series mask arrays. This should match the assumptions (f order, etc) in RnnOutputLayer This reshapes from [X,1] to [X,1,1,1] suitable for per-example masking in CNNs

Parameters:

timeSeriesMask - Mask array to reshape for CNN per-example masking

Returns:

Mask array as 4D CNN mask array: [X, 1, 1, 1]

reshapeVectorToTimeSeriesMask

public static INDArray reshapeVectorToTimeSeriesMask(INDArray timeSeriesMaskAsVector,
                                                     int minibatchSize)

Reshape time series mask arrays. This should match the assumptions (f order, etc) in RnnOutputLayer

Parameters:

timeSeriesMaskAsVector - Mask array to reshape to a column vector

Returns:

Mask array as a column vector

reshapeCnnMaskToTimeSeriesMask

public static INDArray reshapeCnnMaskToTimeSeriesMask(INDArray timeSeriesMaskAsCnnMask,
                                                      int minibatchSize)

Reshape CNN-style 4d mask array of shape [seqLength*minibatch,1,1,1] to time series mask [mb,seqLength] This should match the assumptions (f order, etc) in RnnOutputLayer

Parameters:

timeSeriesMaskAsCnnMask - Mask array to reshape

Returns:

Sequence mask array - [minibatch, sequenceLength]

reshapePerOutputTimeSeriesMaskTo2d

public static INDArray reshapePerOutputTimeSeriesMaskTo2d(INDArray perOutputTimeSeriesMask)

reshapePerOutputTimeSeriesMaskTo2d

public static INDArray reshapePerOutputTimeSeriesMaskTo2d(INDArray perOutputTimeSeriesMask,
                                                          LayerWorkspaceMgr workspaceMgr,
                                                          ArrayType arrayType)

reshape3dTo2d

public static INDArray reshape3dTo2d(INDArray in)

reshape3dTo2d

public static INDArray reshape3dTo2d(INDArray in,
                                     LayerWorkspaceMgr workspaceMgr,
                                     ArrayType arrayType)

reshape2dTo3d

public static INDArray reshape2dTo3d(INDArray in,
                                     int miniBatchSize)

reshape2dTo3d

public static INDArray reshape2dTo3d(INDArray in,
                                     long miniBatchSize,
                                     LayerWorkspaceMgr workspaceMgr,
                                     ArrayType arrayType)

reverseTimeSeries

public static INDArray reverseTimeSeries(INDArray in)

Reverse an input time series along the time dimension

Parameters:

in - Input activations to reverse, with shape [minibatch, size, timeSeriesLength]

Returns:

Reversed activations

reverseTimeSeries

public static INDArray reverseTimeSeries(INDArray in,
                                         LayerWorkspaceMgr workspaceMgr,
                                         ArrayType arrayType,
                                         RNNFormat dataFormat)

reverseTimeSeries

public static INDArray reverseTimeSeries(INDArray in,
                                         LayerWorkspaceMgr workspaceMgr,
                                         ArrayType arrayType)

Reverse an input time series along the time dimension

Parameters:

in - Input activations to reverse, with shape [minibatch, size, timeSeriesLength]

Returns:

Reversed activations

reverseTimeSeriesMask

public static INDArray reverseTimeSeriesMask(INDArray mask)

Reverse a (per time step) time series mask, with shape [minibatch, timeSeriesLength]

Parameters:

mask - Mask to reverse along time dimension

Returns:

Mask after reversing

reverseTimeSeriesMask

public static INDArray reverseTimeSeriesMask(INDArray mask,
                                             LayerWorkspaceMgr workspaceMgr,
                                             ArrayType arrayType)

Reverse a (per time step) time series mask, with shape [minibatch, timeSeriesLength]

Parameters:

mask - Mask to reverse along time dimension

Returns:

Mask after reversing

pullLastTimeSteps

public static Pair<INDArray,int[]> pullLastTimeSteps(INDArray pullFrom,
                                                     INDArray mask)

Extract out the last time steps (2d array from 3d array input) accounting for the mask layer, if present.

Parameters:

pullFrom - Input time series array (rank 3) to pull the last time steps from

mask - Mask array (rank 2). May be null

Returns:

2d array of the last time steps

pullLastTimeSteps

public static Pair<INDArray,int[]> pullLastTimeSteps(INDArray pullFrom,
                                                     INDArray mask,
                                                     LayerWorkspaceMgr workspaceMgr,
                                                     ArrayType arrayType)

Extract out the last time steps (2d array from 3d array input) accounting for the mask layer, if present.

Parameters:

pullFrom - Input time series array (rank 3) to pull the last time steps from

mask - Mask array (rank 2). May be null

Returns:

2d array of the last time steps

getFormatFromRnnLayer

public static RNNFormat getFormatFromRnnLayer(Layer layer)

Get the RNNFormat from the RNN layer, accounting for the presence of wrapper layers like Bidirectional, LastTimeStep, etc

Parameters:

layer - Layer to get the RNNFormat from