com.intel.analytics.bigdl.nn.MultiLabelMarginCriterion.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.TensorCriterion
import com.intel.analytics.bigdl.tensor.Tensor
import com.intel.analytics.bigdl.tensor.TensorNumericMath.TensorNumeric
import scala.reflect.ClassTag
/**
* Creates a criterion that optimizes a multi-class multi-classification hinge loss
* (margin-based loss) between input x and output y (which is a Tensor of target class indices)
*
* @param sizeAverage
*/
@SerialVersionUID(9034717449427139574L)
class MultiLabelMarginCriterion[@specialized(Float, Double) T: ClassTag]
(val sizeAverage: Boolean = true)(implicit ev: TensorNumeric[T]) extends TensorCriterion[T] {
@transient
private var isTarget: Tensor[T] = null
override def updateOutput(input: Tensor[T], target: Tensor[T]): T = {
if (null == isTarget) isTarget = Tensor[T]()
require(input.nDimension() == 1 || input.nDimension() == 2,
"MultiLabelMarginCriterion: " + ErrorInfo.constrainInputAsVectorOrBatch +
s"input dimension ${input.nDimension()}")
val (nframe, dim) = if (input.nDimension() == 1) {
require(target.nDimension() == 1 && target.size(1) == input.size(1),
"MultiLabelMarginCriterion: " + ErrorInfo.constrainInputSizeSameAsTarget +
s"target dimension ${target.nDimension()}, " +
s"target size ${target.size(1)}, input size ${input.size()}")
(1, input.size(1))
} else {
require(target.nDimension() == 2 && target.size(1) == input.size(1) && target.size(2)
== input.size(2), "MultiLabelMarginCriterion: " + ErrorInfo.constrainInputSizeSameAsTarget)
(input.size(1), input.size(2))
}
require(ev.isGreaterEq(target.min(), ev.zero) &&
ev.isGreaterEq(ev.fromType(dim), target.max()), "MultiLabelMarginCriterion: " +
s"target out of range, target min should be greater than or equal to zero, but get " +
s"${target.min()}, target max should be less than or equal to $dim, but get ${target.max()}")
val _target = target.contiguous()
val _input = input.contiguous()
val input_data = _input.storage().array()
val target_data = _target.storage().array()
val input_offset = _input.storageOffset() - 1
val target_offset = _target.storageOffset() - 1
isTarget.resizeAs(target).zero()
val isTarget_data = isTarget.storage().array()
var sum: T = ev.fromType(0)
var t = 0
var n = 0
while (t < nframe) {
var ddt = 0
var dt = 0
while (ddt < dim) {
val target_idx = ev.toType[Int](target_data(n + ddt + target_offset)) - 1
if(target_idx >= 0) {
isTarget_data(n + target_idx) = ev.fromType(1)
ddt += 1
} else {
ddt = dim
}
}
while (dt < dim) {
val target_idx = ev.toType[Int](target_data(n + dt + target_offset)) - 1
if (target_idx >= 0) {
val input_target = input_data(n + target_idx + input_offset)
var d = 0
while (d < dim) {
if (isTarget_data(n + d) == 0) {
val z = ev.plus(ev.minus(ev.fromType(1), input_target),
input_data(n + d + input_offset))
if (ev.isGreater(z, ev.fromType(0))) sum = ev.plus(sum, z)
}
d += 1
}
dt += 1
} else {
dt = dim
}
}
n += dim
t += 1
}
sum = ev.divide(sum, ev.fromType(dim))
if (sizeAverage) sum = ev.divide(sum, ev.fromType(nframe))
sum
}
override def updateGradInput(input: Tensor[T], target: Tensor[T]): Tensor[T] = {
require(input.nDimension() == 1 || input.nDimension() == 2,
"MultiLabelMarginCriterion: " + ErrorInfo.constrainInputAsVectorOrBatch +
s"input dimension ${input.nDimension()}")
val (nframe, dim) = if (input.nDimension() == 1) {
require(target.nDimension() == 1 && target.size(1) == input.size(1),
"MultiLabelMarginCriterion: " + ErrorInfo.constrainInputSizeSameAsTarget +
s"target dimension ${target.nDimension()}" +
s"target size ${target.size(1)} input size ${input.size(1)}")
require(isTarget.nDimension() == 1 && isTarget.size(1) == input.size(1),
"MultiLabelMarginCriterion: inconsistent isTarget size" +
s"isTarget dimension ${isTarget.size(1)}" +
s"isTarget size ${isTarget.size(1)} input size ${input.size(1)}")
(1, input.size(1))
} else {
require(target.nDimension() == 2 && target.size(1) == input.size(1) && target.size(2)
== input.size(2), "MultiLabelMarginCriterion: " + ErrorInfo.constrainInputSizeSameAsTarget +
s"target dimension ${target.nDimension()} " +
s"target size(${target.size(1)},${target.size(2)})" +
s"input size(${input.size(1)},${input.size(2)})")
require(isTarget.nDimension() == 2 && isTarget.size(1) == input.size(1) &&
isTarget.size(2) == input.size(2), "MultiLabelMarginCriterion: inconsistent isTarget size" +
s"isTarget dimension ${isTarget.nDimension()}" +
s"isTarget size(${isTarget.size(1)},${isTarget.size(2)})" +
s"input size(${input.size(1)},${input.size(2)})")
(input.size(1), input.size(2))
}
require(ev.isGreaterEq(target.min(), ev.zero) &&
ev.isGreaterEq(ev.fromType(dim), target.max()), "MultiLabelMarginCriterion: " +
s"target out of range, target min should be greater than or equal to zero, but get " +
s"${target.min()}, target max should be less than or equal to $dim, but get ${target.max()}")
require(ev.isGreaterEq(isTarget.min(), ev.zero) &&
ev.isGreaterEq(ev.fromType(dim), isTarget.max()), "MultiLabelMarginCriterion: " +
"target out of range")
val _target = target.contiguous()
val _input = input.contiguous()
val _isTarget = isTarget.contiguous()
val input_data = _input.storage().array()
val target_data = _target.storage().array()
val isTarget_data = _isTarget.storage().array()
val input_offset = _input.storageOffset() - 1
val target_offset = _target.storageOffset() - 1
val isTarget_offset = _isTarget.storageOffset() - 1
val g = ev.fromType(if (sizeAverage) 1.0/(nframe*dim) else 1.0/(dim))
gradInput.resizeAs(input).zero()
val gradInput_data = gradInput.storage().array()
var t = 0
var n = 0
while (t < nframe) {
var dt = 0
while (dt < dim) {
val target_idx = ev.toType[Int](target_data(n + dt + target_offset)) - 1
if (target_idx >= 0) {
val input_target = input_data(n + target_idx + input_offset)
var d = 0
while (d < dim) {
if (isTarget_data(n + d + isTarget_offset) == 0) {
val z = ev.plus(ev.minus(ev.fromType(1), input_target),
input_data(d + n + input_offset))
if (ev.isGreater(z, ev.fromType(0))) {
gradInput_data(target_idx + n) = ev.minus(gradInput_data(target_idx + n), g)
gradInput_data(d + n) = ev.plus(gradInput_data(d + n), g)
}
}
d += 1
}
dt += 1
} else {
dt = dim
}
}
n += dim
t += 1
}
gradInput
}
override def toString(): String = {
s"nn.MultiLabelMarginCriterion($sizeAverage)"
}
override def canEqual(other: Any): Boolean = other.isInstanceOf[MultiLabelMarginCriterion[T]]
override def equals(other: Any): Boolean = other match {
case that: MultiLabelMarginCriterion[T] =>
super.equals(that) &&
(that canEqual this) &&
sizeAverage == that.sizeAverage
case _ => false
}
override def hashCode(): Int = {
def getHashCode(a: Any): Int = if (a == null) 0 else a.hashCode()
val state = Seq(super.hashCode(), sizeAverage)
state.map(getHashCode).foldLeft(0)((a, b) => 31 * a + b)
}
}
object MultiLabelMarginCriterion {
def apply[@specialized(Float, Double) T: ClassTag](
sizeAverage: Boolean = true)(implicit ev: TensorNumeric[T]) : MultiLabelMarginCriterion[T] = {
new MultiLabelMarginCriterion[T](sizeAverage)
}
}