public <T extends Number> TopK<T> topK(Operand<T> input, Operand<Integer> k, TopK.Options... options)
TopK
operationinput
- 1-D or higher with last dimension at least `k`.k
- 0-D. Number of top elements to look for along the last dimension (along eachoptions
- carries optional attributes valuesTopK
public <T extends Number> DepthwiseConv2dNative<T> depthwiseConv2dNative(Operand<T> input, Operand<T> filter, List<Long> strides, String padding, DepthwiseConv2dNative.Options... options)
DepthwiseConv2dNative
operationinput
- filter
- strides
- 1-D of length 4. The stride of the sliding window for each dimensionpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesDepthwiseConv2dNative
public <T> QuantizedInstanceNorm<T> quantizedInstanceNorm(Operand<T> x, Operand<Float> xMin, Operand<Float> xMax, QuantizedInstanceNorm.Options... options)
QuantizedInstanceNorm
operationx
- A 4D input Tensor.xMin
- The value represented by the lowest quantized input.xMax
- The value represented by the highest quantized input.options
- carries optional attributes valuesQuantizedInstanceNorm
public <T,U extends Number> SpaceToBatch<T> spaceToBatch(Operand<T> input, Operand<U> paddings, Long blockSize)
SpaceToBatch
operationinput
- 4-D with shape `[batch, height, width, depth]`.paddings
- 2-D tensor of non-negative integers with shape `[2, 2]`. It specifiesblockSize
- SpaceToBatch
public <T> DepthToSpace<T> depthToSpace(Operand<T> input, Long blockSize, DepthToSpace.Options... options)
DepthToSpace
operationinput
- blockSize
- The size of the spatial block, same as in Space2Depth.options
- carries optional attributes valuesDepthToSpace
public <T extends Number> Softsign<T> softsign(Operand<T> features)
Softsign
operationfeatures
- Softsign
public <T extends Number> Conv3d<T> conv3d(Operand<T> input, Operand<T> filter, List<Long> strides, String padding, Conv3d.Options... options)
Conv3d
operationinput
- Shape `[batch, in_depth, in_height, in_width, in_channels]`.filter
- Shape `[filter_depth, filter_height, filter_width, in_channels,strides
- 1-D tensor of length 5. The stride of the sliding window for eachpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesConv3d
public <U,T> QuantizedReluX<U> quantizedReluX(Operand<T> features, Operand<Float> maxValue, Operand<Float> minFeatures, Operand<Float> maxFeatures, Class<U> outType)
QuantizedReluX
operationfeatures
- maxValue
- minFeatures
- The float value that the lowest quantized value represents.maxFeatures
- The float value that the highest quantized value represents.outType
- QuantizedReluX
public <T extends Number> MaxPoolWithArgmax<T,Long> maxPoolWithArgmax(Operand<T> input, List<Long> ksize, List<Long> strides, String padding, MaxPoolWithArgmax.Options... options)
MaxPoolWithArgmax
operationinput
- 4-D with shape `[batch, height, width, channels]`. Input to pool over.ksize
- The size of the window for each dimension of the input tensor.strides
- The stride of the sliding window for each dimension of thepadding
- The type of padding algorithm to use.options
- carries optional attributes valuesMaxPoolWithArgmax
public <T extends Number> Conv2dBackpropInput<T> conv2dBackpropInput(Operand<Integer> inputSizes, Operand<T> filter, Operand<T> outBackprop, List<Long> strides, String padding, Conv2dBackpropInput.Options... options)
Conv2dBackpropInput
operationinputSizes
- An integer vector representing the shape of `input`,filter
- 4-D with shapeoutBackprop
- 4-D with shape `[batch, out_height, out_width, out_channels]`.strides
- The stride of the sliding window for each dimension of the inputpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesConv2dBackpropInput
public <U extends Number,T extends Number> MaxPool3dGrad<U> maxPool3dGrad(Operand<T> origInput, Operand<T> origOutput, Operand<U> grad, List<Long> ksize, List<Long> strides, String padding, MaxPool3dGrad.Options... options)
MaxPool3dGrad
operationorigInput
- The original input tensor.origOutput
- The original output tensor.grad
- Output backprop of shape `[batch, depth, rows, cols, channels]`.ksize
- 1-D tensor of length 5. The size of the window for each dimension ofstrides
- 1-D tensor of length 5. The stride of the sliding window for eachpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesMaxPool3dGrad
public ComputeAccidentalHits computeAccidentalHits(Operand<Long> trueClasses, Operand<Long> sampledCandidates, Long numTrue, ComputeAccidentalHits.Options... options)
ComputeAccidentalHits
operationtrueClasses
- The true_classes output of UnpackSparseLabels.sampledCandidates
- The sampled_candidates output of CandidateSampler.numTrue
- Number of true labels per context.options
- carries optional attributes valuesComputeAccidentalHits
public FixedUnigramCandidateSampler fixedUnigramCandidateSampler(Operand<Long> trueClasses, Long numTrue, Long numSampled, Boolean unique, Long rangeMax, FixedUnigramCandidateSampler.Options... options)
FixedUnigramCandidateSampler
operationtrueClasses
- A batch_size * num_true matrix, in which each row contains thenumTrue
- Number of true labels per context.numSampled
- Number of candidates to randomly sample.unique
- If unique is true, we sample with rejection, so that all sampledrangeMax
- The sampler will sample integers from the interval [0, range_max).options
- carries optional attributes valuesFixedUnigramCandidateSampler
public <T extends Number> CudnnRnnCanonicalToParams<T> cudnnRnnCanonicalToParams(Operand<Integer> numLayers, Operand<Integer> numUnits, Operand<Integer> inputSize, Iterable<Operand<T>> weights, Iterable<Operand<T>> biases, CudnnRnnCanonicalToParams.Options... options)
CudnnRnnCanonicalToParams
operationnumLayers
- numUnits
- inputSize
- weights
- biases
- options
- carries optional attributes valuesCudnnRnnCanonicalToParams
public <T extends Number> AvgPool3d<T> avgPool3d(Operand<T> input, List<Long> ksize, List<Long> strides, String padding, AvgPool3d.Options... options)
AvgPool3d
operationinput
- Shape `[batch, depth, rows, cols, channels]` tensor to pool over.ksize
- 1-D tensor of length 5. The size of the window for each dimension ofstrides
- 1-D tensor of length 5. The stride of the sliding window for eachpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesAvgPool3d
public <T extends Number> Relu6<T> relu6(Operand<T> features)
Relu6
operationfeatures
- Relu6
public <T extends Number> Elu<T> elu(Operand<T> features)
Elu
operationfeatures
- Elu
public <T extends Number> CtcLoss<T> ctcLoss(Operand<T> inputs, Operand<Long> labelsIndices, Operand<Integer> labelsValues, Operand<Integer> sequenceLength, CtcLoss.Options... options)
CtcLoss
operationinputs
- 3-D, shape: `(max_time x batch_size x num_classes)`, the logits.labelsIndices
- The indices of a `SparseTensorlabelsValues
- The values (labels) associated with the given batch and time.sequenceLength
- A vector containing sequence lengths (batch).options
- carries optional attributes valuesCtcLoss
public <T extends Number> Conv3dBackpropFilter<T> conv3dBackpropFilter(Operand<T> input, Operand<Integer> filterSizes, Operand<T> outBackprop, List<Long> strides, String padding, Conv3dBackpropFilter.Options... options)
Conv3dBackpropFilter
operationinput
- Shape `[batch, depth, rows, cols, in_channels]`.filterSizes
- An integer vector representing the tensor shape of `filter`,outBackprop
- Backprop signal of shape `[batch, out_depth, out_rows, out_cols,strides
- 1-D tensor of length 5. The stride of the sliding window for eachpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesConv3dBackpropFilter
public <T extends Number> CudnnRnnParamsToCanonical<T> cudnnRnnParamsToCanonical(Operand<Integer> numLayers, Operand<Integer> numUnits, Operand<Integer> inputSize, Operand<T> params, Long numParams, CudnnRnnParamsToCanonical.Options... options)
CudnnRnnParamsToCanonical
operationnumLayers
- numUnits
- inputSize
- params
- numParams
- options
- carries optional attributes valuesCudnnRnnParamsToCanonical
public <T extends Number> FusedPadConv2d<T> fusedPadConv2d(Operand<T> input, Operand<Integer> paddings, Operand<T> filter, String mode, List<Long> strides, String padding)
FusedPadConv2d
operationinput
- 4-D with shape `[batch, in_height, in_width, in_channels]`.paddings
- A two-column matrix specifying the padding sizes. The number offilter
- 4-D with shapemode
- strides
- 1-D of length 4. The stride of the sliding window for each dimensionpadding
- The type of padding algorithm to use.FusedPadConv2d
public <U,T> QuantizedRelu6<U> quantizedRelu6(Operand<T> features, Operand<Float> minFeatures, Operand<Float> maxFeatures, Class<U> outType)
QuantizedRelu6
operationfeatures
- minFeatures
- The float value that the lowest quantized value represents.maxFeatures
- The float value that the highest quantized value represents.outType
- QuantizedRelu6
public <T extends Number> Conv2dBackpropFilter<T> conv2dBackpropFilter(Operand<T> input, Operand<Integer> filterSizes, Operand<T> outBackprop, List<Long> strides, String padding, Conv2dBackpropFilter.Options... options)
Conv2dBackpropFilter
operationinput
- 4-D with shape `[batch, in_height, in_width, in_channels]`.filterSizes
- An integer vector representing the tensor shape of `filter`,outBackprop
- 4-D with shape `[batch, out_height, out_width, out_channels]`.strides
- The stride of the sliding window for each dimension of the inputpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesConv2dBackpropFilter
public <T extends Number> FusedResizeAndPadConv2d<T> fusedResizeAndPadConv2d(Operand<T> input, Operand<Integer> size, Operand<Integer> paddings, Operand<T> filter, String mode, List<Long> strides, String padding, FusedResizeAndPadConv2d.Options... options)
FusedResizeAndPadConv2d
operationinput
- 4-D with shape `[batch, in_height, in_width, in_channels]`.size
- A 1-D int32 Tensor of 2 elements: `new_height, new_width`. Thepaddings
- A two-column matrix specifying the padding sizes. The number offilter
- 4-D with shapemode
- strides
- 1-D of length 4. The stride of the sliding window for each dimensionpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesFusedResizeAndPadConv2d
public <T extends Number> DataFormatDimMap<T> dataFormatDimMap(Operand<T> x, DataFormatDimMap.Options... options)
DataFormatDimMap
operationx
- A Tensor with each element as a dimension index in source data format.options
- carries optional attributes valuesDataFormatDimMap
public <T extends Number> CtcGreedyDecoder<T> ctcGreedyDecoder(Operand<T> inputs, Operand<Integer> sequenceLength, CtcGreedyDecoder.Options... options)
CtcGreedyDecoder
operationinputs
- 3-D, shape: `(max_time x batch_size x num_classes)`, the logits.sequenceLength
- A vector containing sequence lengths, size `(batch_size)`.options
- carries optional attributes valuesCtcGreedyDecoder
public <T extends Number,U extends Number> SparseSoftmaxCrossEntropyWithLogits<T> sparseSoftmaxCrossEntropyWithLogits(Operand<T> features, Operand<U> labels)
SparseSoftmaxCrossEntropyWithLogits
operationfeatures
- batch_size x num_classes matrixlabels
- batch_size vector with values in [0, num_classes).SparseSoftmaxCrossEntropyWithLogits
public <T extends Number,U extends Number> FusedBatchNormGrad<T,U> fusedBatchNormGrad(Operand<T> yBackprop, Operand<T> x, Operand<Float> scale, Operand<U> reserveSpace1, Operand<U> reserveSpace2, FusedBatchNormGrad.Options... options)
FusedBatchNormGrad
operationyBackprop
- A 4D Tensor for the gradient with respect to y.x
- A 4D Tensor for input data.scale
- A 1D Tensor for scaling factor, to scale the normalized x.reserveSpace1
- When is_training is True, a 1D Tensor for the computed batchreserveSpace2
- When is_training is True, a 1D Tensor for the computed batchoptions
- carries optional attributes valuesFusedBatchNormGrad
public <T extends Number> MaxPoolGradGrad<T> maxPoolGradGrad(Operand<T> origInput, Operand<T> origOutput, Operand<T> grad, Operand<Integer> ksize, Operand<Integer> strides, String padding, MaxPoolGradGrad.Options... options)
MaxPoolGradGrad
operationorigInput
- The original input tensor.origOutput
- The original output tensor.grad
- 4-D. Gradients of gradients w.r.t. the input of `max_pool`.ksize
- The size of the window for each dimension of the input tensor.strides
- The stride of the sliding window for each dimension of thepadding
- The type of padding algorithm to use.options
- carries optional attributes valuesMaxPoolGradGrad
public <T> Relu<T> relu(Operand<T> features)
Relu
operationfeatures
- Relu
public LearnedUnigramCandidateSampler learnedUnigramCandidateSampler(Operand<Long> trueClasses, Long numTrue, Long numSampled, Boolean unique, Long rangeMax, LearnedUnigramCandidateSampler.Options... options)
LearnedUnigramCandidateSampler
operationtrueClasses
- A batch_size * num_true matrix, in which each row contains thenumTrue
- Number of true labels per context.numSampled
- Number of candidates to randomly sample.unique
- If unique is true, we sample with rejection, so that all sampledrangeMax
- The sampler will sample integers from the interval [0, range_max).options
- carries optional attributes valuesLearnedUnigramCandidateSampler
public <T> SpaceToDepth<T> spaceToDepth(Operand<T> input, Long blockSize, SpaceToDepth.Options... options)
SpaceToDepth
operationinput
- blockSize
- The size of the spatial block.options
- carries optional attributes valuesSpaceToDepth
public <T> MaxPool<T> maxPool(Operand<T> input, Operand<Integer> ksize, Operand<Integer> strides, String padding, MaxPool.Options... options)
MaxPool
operationinput
- 4-D input to pool over.ksize
- The size of the window for each dimension of the input tensor.strides
- The stride of the sliding window for each dimension of thepadding
- The type of padding algorithm to use.options
- carries optional attributes valuesMaxPool
public <T extends Number,U extends Number> FusedBatchNorm<T,U> fusedBatchNorm(Operand<T> x, Operand<U> scale, Operand<U> offset, Operand<U> mean, Operand<U> variance, FusedBatchNorm.Options... options)
FusedBatchNorm
operationx
- A 4D Tensor for input data.scale
- A 1D Tensor for scaling factor, to scale the normalized x.offset
- A 1D Tensor for offset, to shift to the normalized x.mean
- A 1D Tensor for population mean. Used for inference only;variance
- A 1D Tensor for population variance. Used for inference only;options
- carries optional attributes valuesFusedBatchNorm
public <V,T,U> QuantizedConv2d<V> quantizedConv2d(Operand<T> input, Operand<U> filter, Operand<Float> minInput, Operand<Float> maxInput, Operand<Float> minFilter, Operand<Float> maxFilter, Class<V> outType, List<Long> strides, String padding, QuantizedConv2d.Options... options)
QuantizedConv2d
operationinput
- filter
- filter's input_depth dimension must match input's depth dimensions.minInput
- The float value that the lowest quantized input value represents.maxInput
- The float value that the highest quantized input value represents.minFilter
- The float value that the lowest quantized filter value represents.maxFilter
- The float value that the highest quantized filter value represents.outType
- strides
- The stride of the sliding window for each dimension of the inputpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesQuantizedConv2d
public <T extends Number> LocalResponseNormalization<T> localResponseNormalization(Operand<T> input, LocalResponseNormalization.Options... options)
LocalResponseNormalization
operationinput
- 4-D.options
- carries optional attributes valuesLocalResponseNormalization
public <U,T> QuantizedRelu<U> quantizedRelu(Operand<T> features, Operand<Float> minFeatures, Operand<Float> maxFeatures, Class<U> outType)
QuantizedRelu
operationfeatures
- minFeatures
- The float value that the lowest quantized value represents.maxFeatures
- The float value that the highest quantized value represents.outType
- QuantizedRelu
public <T extends Number> Dilation2d<T> dilation2d(Operand<T> input, Operand<T> filter, List<Long> strides, List<Long> rates, String padding)
Dilation2d
operationinput
- 4-D with shape `[batch, in_height, in_width, depth]`.filter
- 3-D with shape `[filter_height, filter_width, depth]`.strides
- The stride of the sliding window for each dimension of the inputrates
- The input stride for atrous morphological dilation. Must be:padding
- The type of padding algorithm to use.Dilation2d
public <T extends Number> MaxPoolGrad<T> maxPoolGrad(Operand<T> origInput, Operand<T> origOutput, Operand<T> grad, Operand<Integer> ksize, Operand<Integer> strides, String padding, MaxPoolGrad.Options... options)
MaxPoolGrad
operationorigInput
- The original input tensor.origOutput
- The original output tensor.grad
- 4-D. Gradients w.r.t. the output of `max_pool`.ksize
- The size of the window for each dimension of the input tensor.strides
- The stride of the sliding window for each dimension of thepadding
- The type of padding algorithm to use.options
- carries optional attributes valuesMaxPoolGrad
public <T extends Number> NthElement<T> nthElement(Operand<T> input, Operand<Integer> n, NthElement.Options... options)
NthElement
operationinput
- 1-D or higher with last dimension at least `n+1`.n
- 0-D. Position of sorted vector to select along the last dimension (alongoptions
- carries optional attributes valuesNthElement
public <U extends Number,T extends Number> Conv3dBackpropInput<U> conv3dBackpropInput(Operand<T> inputSizes, Operand<U> filter, Operand<U> outBackprop, List<Long> strides, String padding, Conv3dBackpropInput.Options... options)
Conv3dBackpropInput
operationinputSizes
- An integer vector representing the tensor shape of `input`,filter
- Shape `[depth, rows, cols, in_channels, out_channels]`.outBackprop
- Backprop signal of shape `[batch, out_depth, out_rows, out_cols,strides
- 1-D tensor of length 5. The stride of the sliding window for eachpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesConv3dBackpropInput
public <T extends Number> DataFormatVecPermute<T> dataFormatVecPermute(Operand<T> x, DataFormatVecPermute.Options... options)
DataFormatVecPermute
operationx
- Vector of size 4 or Tensor of shape (4, 2) in source data format.options
- carries optional attributes valuesDataFormatVecPermute
public <T> QuantizedMaxPool<T> quantizedMaxPool(Operand<T> input, Operand<Float> minInput, Operand<Float> maxInput, List<Long> ksize, List<Long> strides, String padding)
QuantizedMaxPool
operationinput
- The 4D (batch x rows x cols x depth) Tensor to MaxReduce over.minInput
- The float value that the lowest quantized input value represents.maxInput
- The float value that the highest quantized input value represents.ksize
- The size of the window for each dimension of the input tensor.strides
- The stride of the sliding window for each dimension of the inputpadding
- The type of padding algorithm to use.QuantizedMaxPool
public <T extends Number> Softmax<T> softmax(Operand<T> logits)
Softmax
operationlogits
- 2-D with shape `[batch_size, num_classes]`.Softmax
public <T> BiasAddGrad<T> biasAddGrad(Operand<T> outBackprop, BiasAddGrad.Options... options)
BiasAddGrad
operationoutBackprop
- Any number of dimensions.options
- carries optional attributes valuesBiasAddGrad
public <T> BatchNormWithGlobalNormalizationGrad<T> batchNormWithGlobalNormalizationGrad(Operand<T> t, Operand<T> m, Operand<T> v, Operand<T> gamma, Operand<T> backprop, Float varianceEpsilon, Boolean scaleAfterNormalization)
BatchNormWithGlobalNormalizationGrad
operationt
- A 4D input Tensor.m
- A 1D mean Tensor with size matching the last dimension of t.v
- A 1D variance Tensor with size matching the last dimension of t.gamma
- A 1D gamma Tensor with size matching the last dimension of t.backprop
- 4D backprop Tensor.varianceEpsilon
- A small float number to avoid dividing by 0.scaleAfterNormalization
- A bool indicating whether the resulted tensorBatchNormWithGlobalNormalizationGrad
public <T extends Number,U extends Number> MaxPoolWithArgmax<T,U> maxPoolWithArgmax(Operand<T> input, List<Long> ksize, List<Long> strides, Class<U> Targmax, String padding, MaxPoolWithArgmax.Options... options)
MaxPoolWithArgmax
operationinput
- 4-D with shape `[batch, height, width, channels]`. Input to pool over.ksize
- The size of the window for each dimension of the input tensor.strides
- The stride of the sliding window for each dimension of theTargmax
- padding
- The type of padding algorithm to use.options
- carries optional attributes valuesMaxPoolWithArgmax
public <T extends Number> Dilation2dBackpropFilter<T> dilation2dBackpropFilter(Operand<T> input, Operand<T> filter, Operand<T> outBackprop, List<Long> strides, List<Long> rates, String padding)
Dilation2dBackpropFilter
operationinput
- 4-D with shape `[batch, in_height, in_width, depth]`.filter
- 3-D with shape `[filter_height, filter_width, depth]`.outBackprop
- 4-D with shape `[batch, out_height, out_width, depth]`.strides
- 1-D of length 4. The stride of the sliding window for each dimension ofrates
- 1-D of length 4. The input stride for atrous morphological dilation.padding
- The type of padding algorithm to use.Dilation2dBackpropFilter
public <T extends Number> AvgPool3dGrad<T> avgPool3dGrad(Operand<Integer> origInputShape, Operand<T> grad, List<Long> ksize, List<Long> strides, String padding, AvgPool3dGrad.Options... options)
AvgPool3dGrad
operationorigInputShape
- The original input dimensions.grad
- Output backprop of shape `[batch, depth, rows, cols, channels]`.ksize
- 1-D tensor of length 5. The size of the window for each dimension ofstrides
- 1-D tensor of length 5. The stride of the sliding window for eachpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesAvgPool3dGrad
public <T extends Number> InTopK inTopK(Operand<Float> predictions, Operand<T> targets, Operand<T> k)
InTopK
operationpredictions
- A `batch_size` x `classes` tensor.targets
- A `batch_size` vector of class ids.k
- Number of top elements to look at for computing precision.InTopK
public <U extends Number,T extends Number> CudnnRnnParamsSize<U> cudnnRnnParamsSize(Operand<Integer> numLayers, Operand<Integer> numUnits, Operand<Integer> inputSize, Class<T> T, Class<U> S, CudnnRnnParamsSize.Options... options)
CudnnRnnParamsSize
operationnumLayers
- numUnits
- inputSize
- T
- S
- options
- carries optional attributes valuesCudnnRnnParamsSize
public <T extends Number> Dilation2dBackpropInput<T> dilation2dBackpropInput(Operand<T> input, Operand<T> filter, Operand<T> outBackprop, List<Long> strides, List<Long> rates, String padding)
Dilation2dBackpropInput
operationinput
- 4-D with shape `[batch, in_height, in_width, depth]`.filter
- 3-D with shape `[filter_height, filter_width, depth]`.outBackprop
- 4-D with shape `[batch, out_height, out_width, depth]`.strides
- 1-D of length 4. The stride of the sliding window for each dimension ofrates
- 1-D of length 4. The input stride for atrous morphological dilation.padding
- The type of padding algorithm to use.Dilation2dBackpropInput
public <T> QuantizedAvgPool<T> quantizedAvgPool(Operand<T> input, Operand<Float> minInput, Operand<Float> maxInput, List<Long> ksize, List<Long> strides, String padding)
QuantizedAvgPool
operationinput
- 4-D with shape `[batch, height, width, channels]`.minInput
- The float value that the lowest quantized input value represents.maxInput
- The float value that the highest quantized input value represents.ksize
- The size of the window for each dimension of the input tensor.strides
- The stride of the sliding window for each dimension of the inputpadding
- The type of padding algorithm to use.QuantizedAvgPool
public <T> BatchNormWithGlobalNormalization<T> batchNormWithGlobalNormalization(Operand<T> t, Operand<T> m, Operand<T> v, Operand<T> beta, Operand<T> gamma, Float varianceEpsilon, Boolean scaleAfterNormalization)
BatchNormWithGlobalNormalization
operationt
- A 4D input Tensor.m
- A 1D mean Tensor with size matching the last dimension of t.v
- A 1D variance Tensor with size matching the last dimension of t.beta
- A 1D beta Tensor with size matching the last dimension of t.gamma
- A 1D gamma Tensor with size matching the last dimension of t.varianceEpsilon
- A small float number to avoid dividing by 0.scaleAfterNormalization
- A bool indicating whether the resulted tensorBatchNormWithGlobalNormalization
public <T extends Number> Selu<T> selu(Operand<T> features)
Selu
operationfeatures
- Selu
public <T extends Number> LogSoftmax<T> logSoftmax(Operand<T> logits)
LogSoftmax
operationlogits
- 2-D with shape `[batch_size, num_classes]`.LogSoftmax
public <T extends Number> CtcBeamSearchDecoder<T> ctcBeamSearchDecoder(Operand<T> inputs, Operand<Integer> sequenceLength, Long beamWidth, Long topPaths, CtcBeamSearchDecoder.Options... options)
CtcBeamSearchDecoder
operationinputs
- 3-D, shape: `(max_time x batch_size x num_classes)`, the logits.sequenceLength
- A vector containing sequence lengths, size `(batch)`.beamWidth
- A scalar >= 0 (beam search beam width).topPaths
- A scalar >= 0, <= beam_width (controls output size).options
- carries optional attributes valuesCtcBeamSearchDecoder
public <T extends Number> FractionalAvgPool<T> fractionalAvgPool(Operand<T> value, List<Float> poolingRatio, FractionalAvgPool.Options... options)
FractionalAvgPool
operationvalue
- 4-D with shape `[batch, height, width, channels]`.poolingRatio
- Pooling ratio for each dimension of `value`, currently onlyoptions
- carries optional attributes valuesFractionalAvgPool
public <T extends Number> L2Loss<T> l2Loss(Operand<T> t)
L2Loss
operationt
- Typically 2-D, but may have any dimensions.L2Loss
public <T extends Number> Conv2d<T> conv2d(Operand<T> input, Operand<T> filter, List<Long> strides, String padding, Conv2d.Options... options)
Conv2d
operationinput
- A 4-D tensor. The dimension order is interpreted according to the valuefilter
- A 4-D tensor of shapestrides
- 1-D tensor of length 4. The stride of the sliding window for eachpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesConv2d
public <T extends Number> MaxPool3d<T> maxPool3d(Operand<T> input, List<Long> ksize, List<Long> strides, String padding, MaxPool3d.Options... options)
MaxPool3d
operationinput
- Shape `[batch, depth, rows, cols, channels]` tensor to pool over.ksize
- 1-D tensor of length 5. The size of the window for each dimension ofstrides
- 1-D tensor of length 5. The stride of the sliding window for eachpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesMaxPool3d
public <V,T,U> QuantizedBiasAdd<V> quantizedBiasAdd(Operand<T> input, Operand<U> bias, Operand<Float> minInput, Operand<Float> maxInput, Operand<Float> minBias, Operand<Float> maxBias, Class<V> outType)
QuantizedBiasAdd
operationinput
- bias
- A 1D bias Tensor with size matching the last dimension of 'input'.minInput
- The float value that the lowest quantized input value represents.maxInput
- The float value that the highest quantized input value represents.minBias
- The float value that the lowest quantized bias value represents.maxBias
- The float value that the highest quantized bias value represents.outType
- QuantizedBiasAdd
public <T extends Number> DepthwiseConv2dNativeBackpropInput<T> depthwiseConv2dNativeBackpropInput(Operand<Integer> inputSizes, Operand<T> filter, Operand<T> outBackprop, List<Long> strides, String padding, DepthwiseConv2dNativeBackpropInput.Options... options)
DepthwiseConv2dNativeBackpropInput
operationinputSizes
- An integer vector representing the shape of `input`, basedfilter
- 4-D with shapeoutBackprop
- 4-D with shape based on `data_format`.strides
- The stride of the sliding window for each dimension of the inputpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesDepthwiseConv2dNativeBackpropInput
public <T extends Number> SoftmaxCrossEntropyWithLogits<T> softmaxCrossEntropyWithLogits(Operand<T> features, Operand<T> labels)
SoftmaxCrossEntropyWithLogits
operationfeatures
- batch_size x num_classes matrixlabels
- batch_size x num_classes matrixSoftmaxCrossEntropyWithLogits
public <T extends Number> MaxPool3dGradGrad<T> maxPool3dGradGrad(Operand<T> origInput, Operand<T> origOutput, Operand<T> grad, List<Long> ksize, List<Long> strides, String padding, MaxPool3dGradGrad.Options... options)
MaxPool3dGradGrad
operationorigInput
- The original input tensor.origOutput
- The original output tensor.grad
- Output backprop of shape `[batch, depth, rows, cols, channels]`.ksize
- 1-D tensor of length 5. The size of the window for each dimension ofstrides
- 1-D tensor of length 5. The stride of the sliding window for eachpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesMaxPool3dGradGrad
public <U,T> QuantizedBatchNormWithGlobalNormalization<U> quantizedBatchNormWithGlobalNormalization(Operand<T> t, Operand<Float> tMin, Operand<Float> tMax, Operand<T> m, Operand<Float> mMin, Operand<Float> mMax, Operand<T> v, Operand<Float> vMin, Operand<Float> vMax, Operand<T> beta, Operand<Float> betaMin, Operand<Float> betaMax, Operand<T> gamma, Operand<Float> gammaMin, Operand<Float> gammaMax, Class<U> outType, Float varianceEpsilon, Boolean scaleAfterNormalization)
QuantizedBatchNormWithGlobalNormalization
operationt
- A 4D input Tensor.tMin
- The value represented by the lowest quantized input.tMax
- The value represented by the highest quantized input.m
- A 1D mean Tensor with size matching the last dimension of t.mMin
- The value represented by the lowest quantized mean.mMax
- The value represented by the highest quantized mean.v
- A 1D variance Tensor with size matching the last dimension of t.vMin
- The value represented by the lowest quantized variance.vMax
- The value represented by the highest quantized variance.beta
- A 1D beta Tensor with size matching the last dimension of t.betaMin
- The value represented by the lowest quantized offset.betaMax
- The value represented by the highest quantized offset.gamma
- A 1D gamma Tensor with size matching the last dimension of t.gammaMin
- The value represented by the lowest quantized gamma.gammaMax
- The value represented by the highest quantized gamma.outType
- varianceEpsilon
- A small float number to avoid dividing by 0.scaleAfterNormalization
- A bool indicating whether the resulted tensorQuantizedBatchNormWithGlobalNormalization
public <T extends Number,U extends Number> MaxPoolGradGradWithArgmax<T> maxPoolGradGradWithArgmax(Operand<T> input, Operand<T> grad, Operand<U> argmax, List<Long> ksize, List<Long> strides, String padding, MaxPoolGradGradWithArgmax.Options... options)
MaxPoolGradGradWithArgmax
operationinput
- The original input.grad
- 4-D with shape `[batch, height, width, channels]`. Gradients w.r.t. theargmax
- The indices of the maximum values chosen for each output of `max_pool`.ksize
- The size of the window for each dimension of the input tensor.strides
- The stride of the sliding window for each dimension of thepadding
- The type of padding algorithm to use.options
- carries optional attributes valuesMaxPoolGradGradWithArgmax
public <T extends Number> FractionalMaxPool<T> fractionalMaxPool(Operand<T> value, List<Float> poolingRatio, FractionalMaxPool.Options... options)
FractionalMaxPool
operationvalue
- 4-D with shape `[batch, height, width, channels]`.poolingRatio
- Pooling ratio for each dimension of `value`, currently onlyoptions
- carries optional attributes valuesFractionalMaxPool
public <T extends Number> AvgPool<T> avgPool(Operand<T> value, List<Long> ksize, List<Long> strides, String padding, AvgPool.Options... options)
AvgPool
operationvalue
- 4-D with shape `[batch, height, width, channels]`.ksize
- The size of the sliding window for each dimension of `value`.strides
- The stride of the sliding window for each dimension of `value`.padding
- The type of padding algorithm to use.options
- carries optional attributes valuesAvgPool
public <T> BiasAdd<T> biasAdd(Operand<T> value, Operand<T> bias, BiasAdd.Options... options)
BiasAdd
operationvalue
- Any number of dimensions.bias
- 1-D with size the last dimension of `value`.options
- carries optional attributes valuesBiasAdd
public <T extends Number> DepthwiseConv2dNativeBackpropFilter<T> depthwiseConv2dNativeBackpropFilter(Operand<T> input, Operand<Integer> filterSizes, Operand<T> outBackprop, List<Long> strides, String padding, DepthwiseConv2dNativeBackpropFilter.Options... options)
DepthwiseConv2dNativeBackpropFilter
operationinput
- 4-D with shape based on `data_format`. For example, iffilterSizes
- An integer vector representing the tensor shape of `filter`,outBackprop
- 4-D with shape based on `data_format`.strides
- The stride of the sliding window for each dimension of the inputpadding
- The type of padding algorithm to use.options
- carries optional attributes valuesDepthwiseConv2dNativeBackpropFilter
Copyright © 2022. All rights reserved.