com.intel.analytics.bigdl.utils.Util.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.utils
import java.io._
import com.intel.analytics.bigdl._
import com.intel.analytics.bigdl.nn.Container
import com.intel.analytics.bigdl.nn.tf.Const
import com.intel.analytics.bigdl.optim.DistriOptimizer.{Cache, CacheV1}
import com.intel.analytics.bigdl.tensor.TensorNumericMath.{NumericWildcard, TensorNumeric}
import com.intel.analytics.bigdl.tensor._
import org.apache.commons.lang.SerializationUtils
import org.apache.commons.lang3.SerializationException
import org.apache.spark.rdd.RDD
import scala.reflect.ClassTag
import scala.util.Try
object Util {
def kthLargest(arr: Array[Long], l: Int, r: Int, k: Int): Long = {
if (k == 0) return Long.MaxValue
val pos = randomPartition(arr, l, r)
if (pos-l == k-1) return arr(pos)
if (pos-l > k-1) return kthLargest(arr, l, pos-1, k)
kthLargest(arr, pos + 1, r, k - pos + l - 1)
}
def swap(arr: Array[Long], i: Int, j: Int): Unit = {
val temp = arr(i)
arr(i) = arr(j)
arr(j) = temp
}
private def partition(arr: Array[Long], l: Int, r: Int): Int = {
val x = arr(r)
var i = l
for (j <- l to (r - 1)) {
if (arr(j) > x) {
swap(arr, i, j);
i += 1
}
}
swap(arr, i, r);
i
}
private def randomPartition(arr: Array[Long], l: Int, r: Int): Int = {
val n = r - l + 1;
val pivot = ((Math.random()) % n).toInt;
swap(arr, l + pivot, r);
partition(arr, l, r);
}
private[bigdl] def shift[B](data : Array[B], from : Int, to : Int): Array[B] = {
require(from < data.length && from >= 0, s"invalid from $from array length is ${data.length}")
require(to < data.length && to >= 0, s"invalid to $to array length is ${data.length}")
if (from == to) {
data
} else if (from < to) {
var i = from
while(i < to) {
val tmp = data(i)
data(i) = data(i + 1)
data(i + 1) = tmp
i += 1
}
data
} else {
var i = from
while(i > to) {
val tmp = data(i)
data(i) = data(i - 1)
data(i - 1) = tmp
i -= 1
}
data
}
}
def getAndClearWeightBias[T: ClassTag]
(parameters: (Array[Tensor[T]], Array[Tensor[T]]))(implicit ev: TensorNumeric[T])
: Array[Tensor[T]] = {
if (parameters._1.length != 0) {
var i = 0
val weightsBias = new Array[Tensor[T]](parameters._1.length)
val isQuantized = parameters._1.exists(_.getTensorType == QuantizedType)
val (isCompacted, storage) = if (!isQuantized) {
val storage = Storage(parameters._1(0).storage.array())
(parameters._1.map(_.nElement()).sum == storage.length(), storage)
} else {
(false, null)
}
// get weight and bias
while (i < parameters._1.length) {
if (parameters._1(i) != null) {
val wb = parameters._1(i)
wb.getTensorType match {
case QuantizedType =>
val quantTensor = wb.asInstanceOf[QuantizedTensor[T]]
weightsBias(i) = QuantizedTensor[T](quantTensor.getStorage, quantTensor.maxOfRow,
quantTensor.minOfRow, quantTensor.sumOfRow, quantTensor.size(), quantTensor.params)
case _ =>
weightsBias(i) = if (isCompacted) {
Tensor[T](storage, wb.storageOffset(), wb.size(), wb.stride())
} else {
Tensor[T](Storage(wb.storage().array()), wb.storageOffset(), wb.size(), wb.stride())
}
}
i += 1
}
}
// clear parameters
clearTensor(parameters._1)
clearTensor(parameters._2)
weightsBias
} else {
// just return an empty array when parameters is empty.
Array()
}
}
def getAndClearConsts[T: ClassTag](
model: Container[_, _, T])(implicit ev: TensorNumeric[T]): Map[String, Tensor[_]] = {
val moduleConsts = model.findModules("Const")
.map(_.asInstanceOf[Const[T, _]])
.map(v => (v, v.value.shallowClone()))
moduleConsts.foreach(_._1.value.set())
val result = moduleConsts.map(v => (v._1.getName(), v._2)).toMap[String, Tensor[_]]
require(result.size == moduleConsts.length, s"${model}'s Const node's name is duplicated," +
s"please check your model.")
result
}
def putConsts[T: ClassTag](
model: Container[_, _, T],
consts: Map[String, Tensor[_]])(implicit ev: TensorNumeric[T]) : Unit = {
val moduleConsts = model.findModules("Const")
.map(_.asInstanceOf[Const[T, _]])
moduleConsts.foreach{const =>
val constValue = const.value.asInstanceOf[NumericWildcard]
val constName = const.getName()
constValue.asInstanceOf[Tensor[NumericWildcard]]
.set(consts(constName).asInstanceOf[Tensor[NumericWildcard]])
}
}
def clearTensor[T: ClassTag](tensors: Array[Tensor[T]])
(implicit ev: TensorNumeric[T]): Unit = {
var i = 0
while (i < tensors.length) {
if (tensors(i) != null) {
if (tensors(i).getTensorType == QuantizedType) {
tensors(i).toQuantizedTensor.release()
}
tensors(i).set()
}
i += 1
}
}
def putWeightBias[T: ClassTag](
broadcastWeightBias: Array[Tensor[T]],
localModel: Module[T])(implicit ev: TensorNumeric[T]): Unit = {
val localWeightBias = localModel.parameters()._1
var i = 0
while (i < localWeightBias.length) {
if (localWeightBias(i) != null) {
clearAndSet(localWeightBias(i), broadcastWeightBias(i))
}
i += 1
}
def clearAndSet(old: Tensor[T], other: Tensor[T]): Unit = {
if (old.getTensorType == QuantizedType && other.getTensorType == QuantizedType) {
val quantOld = old.asInstanceOf[QuantizedTensor[T]]
val quantOther = other.asInstanceOf[QuantizedTensor[T]]
if (quantOld.getNativeStorage != quantOther.getNativeStorage) {
quantOld.release()
}
}
old.set(other)
}
}
def initGradWeightBias[T: ClassTag](
broadcastWeightBias: Array[Tensor[T]],
localModel: Module[T])(implicit ev: TensorNumeric[T]): Unit = {
val (localWeightBias, localGradWeightBias) = localModel.parameters()
// init gradient with a compacted storage
val storage = Storage[T](localGradWeightBias.map(_.nElement()).sum)
val isQuantized = broadcastWeightBias.exists(_.getTensorType == QuantizedType)
var i = 0
while (i < localWeightBias.length) {
if (localWeightBias(i) != null) {
val wb = broadcastWeightBias(i)
wb.getTensorType match {
case QuantizedType =>
localGradWeightBias(i).set(Tensor(1))
case _ =>
localGradWeightBias(i).set(storage, wb.storageOffset(), wb.size(), wb.stride())
}
}
i += 1
}
}
/**
* This method is quite like [[org.apache.commons.lang3.SerializationUtils.deserialize]],
* except `resolveClass` method of [[ObjectInputStream]] is overridden,
* which fix potential [[ClassNotFoundException]] caused by uncertain `latestUserDefinedLoader`.
*/
def deserialize[T: ClassTag](objectData: Array[Byte]): T = {
if (objectData == null) {
throw new IllegalArgumentException("The byte[] must not be null")
}
deserialize[T](new ByteArrayInputStream(objectData))
}
/**
* This method is quite like [[org.apache.commons.lang3.SerializationUtils.deserialize]],
* except `resolveClass` method of [[ObjectInputStream]] is overridden,
* which fix potential [[ClassNotFoundException]] caused by uncertain `latestUserDefinedLoader`.
*/
def deserialize[T: ClassTag](inputStream: InputStream): T = {
if (inputStream == null) {
throw new IllegalArgumentException("The InputStream must not be null")
}
var in: ObjectInputStream = null
try {
// stream closed in the finally
in = new ObjectInputStream(inputStream) {
override def resolveClass(desc: ObjectStreamClass): Class[_] = {
Try(Class.forName(desc.getName, false, getClass.getClassLoader)
).getOrElse(super.resolveClass(desc))
}
}
in.readObject().asInstanceOf[T]
} catch {
case ex: ClassCastException => throw new SerializationException(ex)
case ex: ClassNotFoundException => throw new SerializationException(ex)
case ex: IOException => throw new SerializationException(ex)
} finally {
if (in != null) Try(in.close())
}
}
def cloneParameters[T: ClassTag]
(parameters: Array[Tensor[T]])(implicit ev: TensorNumeric[T])
: Array[Tensor[T]] = {
if (parameters != null) {
if (parameters.length != 0) {
var i = 0
val retParams = new Array[Tensor[T]](parameters.length)
val isQuantized = parameters.exists(_.getTensorType == QuantizedType)
val (isCompacted, storage) = if (!isQuantized) {
val storage = Storage(parameters(0).storage.array())
(parameters.map(_.nElement()).sum == storage.length(), storage)
} else {
(false, null)
}
val resultStorage = if (isCompacted) {
val resultStorage = Storage[T](storage.length())
System.arraycopy(storage.array(), parameters(0).storageOffset() - 1,
resultStorage.array(), 0, storage.length())
resultStorage
} else {
null
}
// clone parameters
while (i < parameters.length) {
if (parameters(i) != null) {
val param = parameters(i)
param.getTensorType match {
case QuantizedType =>
val quantizedTensor = param.asInstanceOf[QuantizedTensor[T]]
val storage = new Array[Byte](quantizedTensor.nElement())
System.arraycopy(quantizedTensor.getStorage, 0,
storage, 0, quantizedTensor.nElement())
val sizeLength = quantizedTensor.size.length
val size = new Array[Int](sizeLength)
System.arraycopy(quantizedTensor.size, 0, size, 0, sizeLength)
val params = quantizedTensor.params.copy()
val length = quantizedTensor.maxOfRow.length
val maxOfRow = new Array[T](length)
System.arraycopy(quantizedTensor.maxOfRow, 0, maxOfRow, 0, length)
val minOfRow = new Array[T](length)
System.arraycopy(quantizedTensor.minOfRow, 0, minOfRow, 0, length)
val sumOfRow = new Array[T](length)
System.arraycopy(quantizedTensor.sumOfRow, 0, sumOfRow, 0, length)
retParams(i) = QuantizedTensor[T](storage, maxOfRow, minOfRow, sumOfRow, size,
params)
case _ =>
retParams(i) = if (isCompacted) {
Tensor[T](resultStorage, param.storageOffset(), param.size(), param.stride())
} else {
param.clone()
}
}
}
i += 1
}
retParams
} else {
// just return an empty array when parameters is empty.
Array()
}
} else {
null
}
}
def setExtraParametersFromModelRDD[T: ClassTag]
(models: RDD[Cache[T]], trainingModel: Module[T], maxSize: Int)(implicit ev: TensorNumeric[T])
: Unit = {
if (trainingModel.getExtraParameter() != null && trainingModel.getExtraParameter().length > 0) {
val totalElements = models.map(_.localModels.head.getExtraParameter().map(_.nElement()).
reduce(_ + _)).first()
val extraStates = if (totalElements < maxSize) {
models.map(_.localModels.head.getExtraParameter()).first()
} else {
val individualLength = models.map(_.localModels.head.getExtraParameter().
map(_.nElement())).first()
val extraParamLength = individualLength.length
val extraState = new Array[Tensor[T]](extraParamLength)
(0 until extraParamLength).foreach(i =>
if (individualLength(i) < maxSize) {
extraState(i) = models.map(_.localModels.head.getExtraParameter()(i)).first()
} else {
val numChucks = if (individualLength(i) % maxSize == 0) {
individualLength(i) / maxSize
} else {
individualLength(i) / maxSize + 1
}
val storage = Storage(new Array[T](individualLength(i)))
for (j <- 0 until numChucks) {
val partArray = models.map(_.localModels.head.getExtraParameter()(i).storage().array()
.slice(j * maxSize, math.min(maxSize * (j + 1), individualLength(i)))).first()
System.arraycopy(partArray, 0, storage.array(), j * maxSize, partArray.length)
}
val trainParam = trainingModel.getExtraParameter()(i)
extraState(i) = Tensor(storage, trainParam.storageOffset(),
trainParam.size, trainParam.stride())
}
)
extraState
}
trainingModel.setExtraParameter(extraStates)
}
}
}
© 2015 - 2025 Weber Informatics LLC | Privacy Policy