Class InputType
- java.lang.Object
-
- org.deeplearning4j.nn.conf.inputs.InputType
-
- All Implemented Interfaces:
Serializable
- Direct Known Subclasses:
InputType.InputTypeConvolutional
,InputType.InputTypeConvolutional3D
,InputType.InputTypeConvolutionalFlat
,InputType.InputTypeFeedForward
,InputType.InputTypeRecurrent
public abstract class InputType extends Object implements Serializable
- See Also:
- Serialized Form
-
-
Nested Class Summary
Nested Classes Modifier and Type Class Description static class
InputType.InputTypeConvolutional
static class
InputType.InputTypeConvolutional3D
static class
InputType.InputTypeConvolutionalFlat
static class
InputType.InputTypeFeedForward
static class
InputType.InputTypeRecurrent
static class
InputType.Type
The type of activations in/out of a given GraphVertex
FF: Standard feed-foward (2d minibatch, 1d per example) data
RNN: Recurrent neural network (3d minibatch) time series data
CNN: 2D Convolutional neural network (4d minibatch, [miniBatchSize, channels, height, width]) CNNFlat: Flattened 2D conv net data (2d minibatch, [miniBatchSize, height * width * channels]) CNN3D: 3D convolutional neural network (5d minibatch, [miniBatchSize, channels, height, width, channels])
-
Constructor Summary
Constructors Constructor Description InputType()
-
Method Summary
All Methods Static Methods Instance Methods Abstract Methods Concrete Methods Deprecated Methods Modifier and Type Method Description abstract long
arrayElementsPerExample()
static InputType
convolutional(long height, long width, long depth)
Input type for convolutional (CNN) data, that is 4d with shape [miniBatchSize, channels, height, width].static InputType
convolutional(long height, long width, long depth, CNN2DFormat format)
static InputType
convolutional3D(long depth, long height, long width, long channels)
static InputType
convolutional3D(Convolution3D.DataFormat dataFormat, long depth, long height, long width, long channels)
Input type for 3D convolutional (CNN3D) 5d data:
If NDHWC format [miniBatchSize, depth, height, width, channels]
If NDCWHstatic InputType
convolutionalFlat(long height, long width, long depth)
Input type for convolutional (CNN) data, where the data is in flattened (row vector) format.static InputType
feedForward(long size)
InputType for feed forward network datastatic InputType
feedForward(long size, DataFormat timeDistributedFormat)
static CNN2DFormat
getDefaultCNN2DFormat()
long[]
getShape()
Returns the shape of this InputType without minibatch dimension in the returned arrayabstract long[]
getShape(boolean includeBatchDim)
Returns the shape of this InputTypeabstract InputType.Type
getType()
static InputType
inferInputType(INDArray inputArray)
static InputType[]
inferInputTypes(INDArray... inputArrays)
static InputType
recurrent(long size)
InputType for recurrent neural network (time series) datastatic InputType
recurrent(long size, long timeSeriesLength)
InputType for recurrent neural network (time series) datastatic InputType
recurrent(long size, long timeSeriesLength, RNNFormat format)
static InputType
recurrent(long size, RNNFormat format)
static void
setDefaultCNN2DFormat(CNN2DFormat defaultCNN2DFormat)
abstract String
toString()
-
-
-
Method Detail
-
getDefaultCNN2DFormat
public static CNN2DFormat getDefaultCNN2DFormat()
-
setDefaultCNN2DFormat
public static void setDefaultCNN2DFormat(CNN2DFormat defaultCNN2DFormat)
-
getType
public abstract InputType.Type getType()
-
arrayElementsPerExample
public abstract long arrayElementsPerExample()
-
getShape
public abstract long[] getShape(boolean includeBatchDim)
Returns the shape of this InputType- Parameters:
includeBatchDim
- Whether to include minibatch in the return shape array- Returns:
- int[]
-
getShape
public long[] getShape()
Returns the shape of this InputType without minibatch dimension in the returned array- Returns:
- int[]
-
feedForward
public static InputType feedForward(long size)
InputType for feed forward network data- Parameters:
size
- The size of the activations- Returns:
- InputTypeFeedForward
-
feedForward
public static InputType feedForward(long size, DataFormat timeDistributedFormat)
-
recurrent
public static InputType recurrent(long size)
InputType for recurrent neural network (time series) data- Parameters:
size
- The size of the activations- Returns:
- InputTypeRecurrent
-
recurrent
public static InputType recurrent(long size, long timeSeriesLength)
InputType for recurrent neural network (time series) data- Parameters:
size
- The size of the activationstimeSeriesLength
- Length of the input time series- Returns:
- InputTypeRecurrent
-
convolutional
public static InputType convolutional(long height, long width, long depth)
Input type for convolutional (CNN) data, that is 4d with shape [miniBatchSize, channels, height, width]. For CNN data that has been flattened, useconvolutionalFlat(long, long, long)
- Parameters:
height
- height of the inputwidth
- Width of the inputdepth
- Depth, or number of channels- Returns:
- InputTypeConvolutional
-
convolutional
public static InputType convolutional(long height, long width, long depth, CNN2DFormat format)
-
convolutional3D
@Deprecated public static InputType convolutional3D(long depth, long height, long width, long channels)
Input type for 3D convolutional (CNN3D) data in NDHWC format, that is 5d with shape [miniBatchSize, depth, height, width, channels].- Parameters:
height
- height of the inputwidth
- Width of the inputdepth
- Depth of the inputchannels
- Number of channels of the input- Returns:
- InputTypeConvolutional3D
-
convolutional3D
public static InputType convolutional3D(Convolution3D.DataFormat dataFormat, long depth, long height, long width, long channels)
Input type for 3D convolutional (CNN3D) 5d data:
If NDHWC format [miniBatchSize, depth, height, width, channels]
If NDCWH- Parameters:
height
- height of the inputwidth
- Width of the inputdepth
- Depth of the inputchannels
- Number of channels of the input- Returns:
- InputTypeConvolutional3D
-
convolutionalFlat
public static InputType convolutionalFlat(long height, long width, long depth)
Input type for convolutional (CNN) data, where the data is in flattened (row vector) format. Expect data with shape [miniBatchSize, height * width * channels]. For CNN data in 4d format, useconvolutional(long, long, long)
- Parameters:
height
- Height of the (unflattened) data represented by this input typewidth
- Width of the (unflattened) data represented by this input typedepth
- Depth of the (unflattened) data represented by this input type- Returns:
- InputTypeConvolutionalFlat
-
-