com.intel.analytics.bigdl.nn.LocallyConnected1D.scala Maven / Gradle / Ivy
The newest version!
/*
* Copyright 2016 The BigDL Authors.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package com.intel.analytics.bigdl.nn
import com.intel.analytics.bigdl.nn.abstractnn.{Initializable, TensorModule}
import com.intel.analytics.bigdl.optim.Regularizer
import com.intel.analytics.bigdl.tensor.Tensor
import com.intel.analytics.bigdl.tensor.TensorNumericMath.TensorNumeric
import com.intel.analytics.bigdl.utils.{Engine, T, Table}
import scala.concurrent.Future
import scala.reflect.ClassTag
/*
* The `LocallyConnected1D` layer works similarly to
* the `TemporalConvolution` layer, except that weights are unshared,
* that is, a different set of filters is applied at each different patch
* of the input.
* The input tensor in `forward(input)` is expected to be a 2D tensor
* (`nInputFrame` x `inputFrameSize`) or a 3D tensor
* (`nBatchFrame` x `nInputFrame` x `inputFrameSize`).
*
* @param nInputFrame the input frame channel
* @param inputFrameSize The input frame size expected in sequences given into `forward()`.
* @param outputFrameSize The output frame size the convolution layer will produce.
* @param kernelW The kernel width of the convolution
* @param strideW The step of the convolution in the width dimension.
* @param propagateBack Whether propagate gradient back, default is true.
* @param wRegularizer instance of [[Regularizer]]
* (eg. L1 or L2 regularization), applied to the input weights matrices.
* @param bRegularizer instance of [[Regularizer]]
* applied to the bias.
* @param initWeight Initial weight
* @param initBias Initial bias
* @param initGradWeight Initial gradient weight
* @param initGradBias Initial gradient bias
* @tparam T The numeric type in the criterion, usually which are [[Float]] or [[Double]]
*/
class LocallyConnected1D[T: ClassTag](val nInputFrame: Int,
val inputFrameSize: Int,
val outputFrameSize: Int,
val kernelW: Int,
val strideW: Int = 1,
val propagateBack: Boolean = true,
var wRegularizer: Regularizer[T] = null,
var bRegularizer: Regularizer[T] = null,
val initWeight: Tensor[T] = null,
val initBias: Tensor[T] = null,
val initGradWeight: Tensor[T] = null,
val initGradBias: Tensor[T] = null
)(implicit ev: TensorNumeric[T])
extends TensorModule[T] with Initializable {
val nOutputFrame = (nInputFrame - kernelW) / strideW + 1
val weight: Tensor[T] = if (initWeight != null) {
initWeight
} else {
Tensor[T](nOutputFrame, outputFrameSize, inputFrameSize * kernelW)
}
val bias: Tensor[T] = if (initBias != null) {
initBias
} else {
Tensor[T](nOutputFrame, outputFrameSize)
}
val gradWeight: Tensor[T] = if (initGradWeight != null) {
initGradWeight
} else {
Tensor[T](nOutputFrame, outputFrameSize, inputFrameSize * kernelW)
}
val gradBias: Tensor[T] = if (initGradBias != null) {
initGradBias
} else {
Tensor[T](nOutputFrame, outputFrameSize)
}
@transient protected var inputWindow: Tensor[T] = _
@transient protected var outputWindow: Tensor[T] = _
@transient protected var weightWindow: Tensor[T] = _
@transient protected var biasWindow: Tensor[T] = _
@transient protected var gradInputWindow: Tensor[T] = _
@transient protected var gradOutputWindow: Tensor[T] = _
@transient protected var gradWeightWindow: Tensor[T] = _
{
val stdv = 1.0 / math.sqrt(kernelW * inputFrameSize)
val wInit: InitializationMethod = RandomUniform(-stdv, stdv)
val bInit: InitializationMethod = RandomUniform(-stdv, stdv)
setInitMethod(wInit, bInit)
}
@transient
protected var results: Array[Future[Unit]] = _
override def reset(): Unit = {
if (initWeight == null) {
weightInitMethod.init(weight, VariableFormat.OUT_IN)
}
if (initBias == null) {
biasInitMethod.init(bias, VariableFormat.ONE_D)
}
zeroGradParameters()
}
def reshapeInput(input: Tensor[T]): Tensor[T] = {
if (input.dim() == 2) {
input.reshape(Array(1, input.size(1), input.size(2)))
} else {
input
}
}
def reshapeOutput(input: Tensor[T], output: Tensor[T]): Tensor[T] = {
if (input.dim() == 2) {
output.reshape(Array(output.size(2), output.size(3)))
} else {
output
}
}
override def updateOutput(_input: Tensor[T]): Tensor[T] = {
// Require input of 2 dimensions or 3 dimensions
// 2d input format: time x feature
// 3d input format: batch x time x feature
require(_input.dim() == 2 || _input.dim() == 3,
"LocallyConvolution1D: 2D or 3D(batch mode) tensor expected for input, " +
s"but got ${_input.dim()}")
// Require input to be contiguous
require(_input.isContiguous())
val input = reshapeInput(_input)
var dimSeq = input.dim() - 1 // 1
var dimFeat = dimSeq + 1 // 2
val nInputFrame = input.size(dimSeq) // 10
var nOutputFrame = (nInputFrame - kernelW) / strideW + 1 // (10 -3)/1 +1 = 8
if (inputWindow == null) inputWindow = Tensor[T]()
if (outputWindow == null) outputWindow = Tensor[T]()
if (weightWindow == null) weightWindow = Tensor[T]()
if (biasWindow == null) biasWindow = Tensor[T]()
// Shape check on input with inputFrameSize and kernelW
require(input.size(dimFeat) == inputFrameSize, "Invalid input frame size. Got: " +
s"${input.size(dimFeat)}, Expected: $inputFrameSize")
require(nOutputFrame >= 1, "Input sequence smaller than kernel size. Got: " +
s"$nInputFrame, Expected: $kernelW")
val batchSize = input.size(1)
val pageSize = weight.size(2) * weight.size(3)
output.resize(batchSize, nOutputFrame, outputFrameSize)
if (results == null || results.length != batchSize) {
results = new Array[Future[Unit]](batchSize)
}
var i = 0
while (i < batchSize) {
results(i) = Engine.model.invoke(() => {
val inputSample = input.select(1, i + 1)
val outputSample = output.select(1, i + 1)
// Add bias first
var j = 1
while (j < nOutputFrame) {
biasWindow = bias.select(1, j)
outputWindow = outputSample.select(dimSeq - 1, j) // setup up bias for each ouputframe
outputWindow.copy(biasWindow)
j += 1
}
// Add the convolution part
j = 0
while (j < nOutputFrame) {
inputWindow.set(inputSample.storage(), inputSample.storageOffset() +
j * strideW * input.size(dimFeat),
Array(1, kernelW * input.size(dimFeat)),
Array(1, 1))
outputWindow.set(outputSample.storage(), outputSample.storageOffset() +
j * output.size(dimFeat),
Array(1, output.size(dimFeat)),
Array(1, 1))
val weightT = weightWindow.set(weight.storage(), weight.storageOffset() +
j * pageSize,
Array(output.size(dimFeat), kernelW * input.size(dimFeat)),
Array(kernelW * input.size(dimFeat), 1)
).transpose(1, 2)
outputWindow.addmm(ev.fromType[Int](1), outputWindow,
ev.fromType[Int](1), inputWindow, weightT)
j += 1
}
})
i += 1
}
output = reshapeOutput(_input, output)
output
}
override def updateGradInput(_input: Tensor[T], _gradOutput: Tensor[T]): Tensor[T] = {
// Require input of 2 dimensions or 3 dimensions
// 2d input format: time x feature
// 3d input format: batch x time x feature
require(_input.dim() == 2 || _input.dim() == 3,
"TemporalConvolution: 2D or 3D(batch mode) tensor expected for input, " +
s"but got ${_input.dim()}")
// Require input to be contiguous
require(_input.isContiguous())
val input = reshapeInput(_input)
val gradOutput = reshapeInput(_gradOutput)
val dimSeq = if (input.dim() == 2) 1 else 2
val dimFeat = if (input.dim() == 2) 2 else 3
val nInputFrame = input.size(dimSeq)
var nOutputFrame = (nInputFrame - kernelW) / strideW + 1
if (gradInputWindow == null) gradInputWindow = Tensor[T]()
if (gradOutputWindow == null) gradOutputWindow = Tensor[T]()
if (weightWindow == null) weightWindow = Tensor[T]()
// Shape check on input with inputFrameSize and kernelW
require(input.size(dimFeat) == inputFrameSize, "Invalid input frame size. Got: " +
s"${input.size(dimFeat)}, Expected: $inputFrameSize")
require(nOutputFrame >= 1, "Input sequence smaller than kernel size. Got: " +
s"$nInputFrame, Expected: $kernelW")
gradInput.resizeAs(input)
gradInput.zero()
val batchSize = input.size(1)
val pageSize = weight.size(2) * weight.size(3)
var gradOutputSample = Tensor[T]()
var gradInputSample = Tensor[T]()
var i = 0
while (i < batchSize) {
results(i) = Engine.model.invoke(() => {
gradInputSample = gradInput.select(1, i + 1)
gradOutputSample = gradOutput.select(1, i + 1)
var j = 0
while (j < nOutputFrame) {
gradOutputWindow.set(gradOutputSample.storage(),
gradOutputSample.storageOffset() + j * gradOutput.size(dimFeat),
Array(1, gradOutput.size(dimFeat)),
Array(1, 1))
gradInputWindow.set(gradInputSample.storage(),
gradInputSample.storageOffset() + j * strideW * gradInput.size(dimFeat),
Array(1, kernelW * gradInput.size(dimFeat)),
Array(1, 1))
weightWindow.set(weight.storage(), weight.storageOffset() + j * pageSize,
Array(output.size(dimFeat), kernelW * input.size(dimFeat)),
Array(kernelW * input.size(dimFeat), 1)
)
gradInputWindow.addmm(ev.fromType[Int](1), gradInputWindow,
ev.fromType[Int](1), gradOutputWindow, weightWindow)
j += 1
}
})
i += 1
}
gradInput = reshapeOutput(_gradOutput, gradInput)
gradInput
}
override def accGradParameters(_input: Tensor[T], _gradOutput: Tensor[T]): Unit = {
// Require input of 2 dimensions or 3 dimensions
require(_input.nDimension() == 2 || _input.nDimension() == 3,
"Only support 2D or 3D input, " +
s"input ${_input.nDimension()}")
// Require input to be contiguous
require(_gradOutput.isContiguous())
val input = reshapeInput(_input)
val gradOutput = reshapeInput(_gradOutput)
val dimSeq = if (input.dim() == 2) 1 else 2
val dimFeat = if (input.dim() == 2) 2 else 3
val nInputFrame = input.size(dimSeq)
var nOutputFrame = (nInputFrame - kernelW) / strideW + 1
if (gradOutputWindow == null) gradOutputWindow = Tensor[T]()
if (inputWindow == null) inputWindow = Tensor[T]()
if (gradWeightWindow == null) gradWeightWindow = Tensor[T]()
if (biasWindow == null) biasWindow = Tensor[T]()
val batchSize = input.size(1)
var gradOutputSample = Tensor[T]()
var inputSample = Tensor[T]()
var i = 0
while (i < batchSize) {
results(i) = Engine.model.invoke(() => {
gradOutputSample = gradOutput.select(1, i + 1)
inputSample = input.select(1, i + 1)
var j = 0
while (j < nOutputFrame) {
biasWindow.set(gradBias.storage(),
gradBias.storageOffset() + j * gradOutput.size(dimFeat),
Array(1, gradOutput.size(dimFeat)),
Array(1, 1))
gradOutputWindow.set(gradOutputSample.select(1, j + 1))
biasWindow.add(biasWindow, ev.fromType[Double](scaleB), gradOutputWindow)
j += 1
}
j = 0
while (j < nOutputFrame) {
inputWindow.set(inputSample.storage(), inputSample.storageOffset() +
j * strideW * input.size(dimFeat),
Array(1, kernelW * input.size(dimFeat)),
Array(1, 1))
gradOutputWindow.set(gradOutputSample.storage(),
gradOutputSample.storageOffset() + j * gradOutput.size(dimFeat),
Array(1, gradOutput.size(dimFeat)),
Array(1, 1))
val gradOutputWindowT = gradOutputWindow.transpose(1, 2)
val pageSize = weight.size(2) * weight.size(3)
gradWeightWindow.set(gradWeight.storage(), gradWeight.storageOffset() +
j * pageSize,
Array(gradOutput.size(dimFeat), kernelW * input.size(dimFeat)),
Array(kernelW * input.size(dimFeat), 1))
gradWeightWindow.addmm(ev.fromType[Int](1), gradWeightWindow,
ev.fromType[Double](scaleW), gradOutputWindowT, inputWindow)
j += 1
}
})
i += 1
}
if (null != wRegularizer) {
wRegularizer.accRegularization(weight, gradWeight, scaleW)
}
if (null != bRegularizer) {
bRegularizer.accRegularization(bias, gradBias, scaleB)
}
}
override def parameters(): (Array[Tensor[T]], Array[Tensor[T]]) = {
(Array(this.weight, this.bias), Array(this.gradWeight, this.gradBias))
}
override def equals(obj: Any): Boolean = {
if (!super.equals(obj)) {
return false
}
if (!obj.isInstanceOf[TemporalConvolution[T]]) {
return false
}
val other = obj.asInstanceOf[TemporalConvolution[T]]
if (this.eq(other)) {
return true
}
inputFrameSize == other.inputFrameSize &&
outputFrameSize == other.outputFrameSize &&
kernelW == other.kernelW &&
strideW == other.strideW &&
propagateBack == other.propagateBack &&
weight == other.weight &&
bias == other.bias &&
gradWeight == other.gradWeight &&
gradBias == other.gradBias
}
override def hashCode(): Int = {
val seed = 37
var hash = super.hashCode()
hash = hash * seed + inputFrameSize.hashCode()
hash = hash * seed + outputFrameSize.hashCode()
hash = hash * seed + kernelW.hashCode()
hash = hash * seed + strideW.hashCode()
hash = hash * seed + weight.hashCode()
hash = hash * seed + bias.hashCode()
hash = hash * seed + gradWeight.hashCode()
hash = hash * seed + gradBias.hashCode()
hash
}
override def clearState(): this.type = {
super.clearState()
this
}
override def toString(): String = {
s"nn.TemporalConvolution($inputFrameSize -> $outputFrameSize, $kernelW x $strideW)"
}
}
object LocallyConnected1D {
def apply[@specialized(Float, Double) T: ClassTag](
nInputFrame: Int,
inputFrameSize: Int,
outputFrameSize: Int,
kernelW: Int,
strideW: Int = 1,
propagateBack: Boolean = true,
wRegularizer: Regularizer[T] = null,
bRegularizer: Regularizer[T] = null,
initWeight: Tensor[T] = null,
initBias: Tensor[T] = null,
initGradWeight: Tensor[T] = null,
initGradBias: Tensor[T] = null
)(implicit ev: TensorNumeric[T]):
LocallyConnected1D[T] = {
new LocallyConnected1D[T](nInputFrame, inputFrameSize, outputFrameSize, kernelW,
strideW, propagateBack,
wRegularizer, bRegularizer, initWeight, initBias, initGradWeight, initGradBias)
}
}