572 lines
25 KiB
C++
572 lines
25 KiB
C++
/* ******************************************************************************
|
|
*
|
|
*
|
|
* This program and the accompanying materials are made available under the
|
|
* terms of the Apache License, Version 2.0 which is available at
|
|
* https://www.apache.org/licenses/LICENSE-2.0.
|
|
*
|
|
* See the NOTICE file distributed with this work for additional
|
|
* information regarding copyright ownership.
|
|
* 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.
|
|
*
|
|
* SPDX-License-Identifier: Apache-2.0
|
|
******************************************************************************/
|
|
|
|
//
|
|
// @author Yurii Shyrma (iuriish@yahoo.com), created on 25.11.2017.
|
|
//
|
|
|
|
#include <system/op_boilerplate.h>
|
|
#if NOT_EXCLUDED(OP_softmax_cross_entropy_loss)
|
|
|
|
#include <ops/declarable/CustomOperations.h>
|
|
|
|
namespace sd {
|
|
namespace ops {
|
|
|
|
//////////////////////////////////////////////////////////////////////////
|
|
CUSTOM_OP_IMPL(softmax_cross_entropy_loss, 3, 1, false, 1, 1) {
|
|
auto logits = INPUT_VARIABLE(0);
|
|
auto weights = INPUT_VARIABLE(1);
|
|
auto labels = INPUT_VARIABLE(2);
|
|
auto output = OUTPUT_VARIABLE(0);
|
|
|
|
int reductionMode =
|
|
INT_ARG(0); // 0 - "none"; 1 - "weighted_sum"; 2 - "weighted_mean"; 3 - "weighted_sum_by_nonzero_weights"
|
|
double labelsSmoothing = T_ARG(0);
|
|
|
|
// input validation
|
|
REQUIRE_TRUE(labels->isSameShape(logits), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS OP: labels and logits arrays must have the same shapes, but got %s and %s "
|
|
"correspondingly !",
|
|
ShapeUtils::shapeAsString(labels).c_str(), ShapeUtils::shapeAsString(logits).c_str());
|
|
// only 4 possible reduction modes exist
|
|
REQUIRE_TRUE(reductionMode == 0 || reductionMode == 1 || reductionMode == 2 || reductionMode == 3, 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS OP: reduction mode value is not acceptable, possible values are 0, 1, 2, 3, "
|
|
"but got %i instead!",
|
|
reductionMode);
|
|
// smoothing is possible for rank of logits/labels > 1
|
|
REQUIRE_TRUE(labels->rankOf() > 1 || (labels->rankOf() == 1 && labelsSmoothing == 0.), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS OP: smoothing is not possible when rank of labels/ logits = 1 !");
|
|
|
|
if (!output->isScalar()) {
|
|
// weights array can be single scalar or has the same shape as output, and must be broadcastable to output shape
|
|
REQUIRE_TRUE(weights->isScalar() || weights->rankOf() == output->rankOf(), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS OP: weights array should be scalar or have the same rank as output array, "
|
|
"but got %i and %i correspondingly!",
|
|
weights->rankOf(), output->rankOf());
|
|
// check whether broadcast operation is possible for weights array
|
|
REQUIRE_TRUE(weights->isScalar() || ShapeUtils::areShapesBroadcastable(*weights, *output), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS OP: shapes of weights and output arrays should be broadcastable, but got "
|
|
"weights = %s and output = %s instead!",
|
|
ShapeUtils::shapeAsString(weights).c_str(), ShapeUtils::shapeAsString(labels).c_str());
|
|
}
|
|
|
|
// If label_smoothing is nonzero, smooth the labels towards 1/num_classes: new_onehot_labels = onehot_labels * (1 -
|
|
// label_smoothing) + label_smoothing / num_classes num_classes = labels->sizeAt(1)
|
|
NDArray* cLabels = labels->cast(weights->dataType()); // cast() already returns NDArray*
|
|
NDArray* newLabels = cLabels;
|
|
if (labelsSmoothing != 0.) {
|
|
newLabels = new NDArray(cLabels);
|
|
NDArray* term1 = (1.f - labelsSmoothing) * (*cLabels);
|
|
NDArray* term2 = (*term1) + (labelsSmoothing / cLabels->sizeAt(1));
|
|
delete term1;
|
|
newLabels->assign(term2);
|
|
delete term2;
|
|
}
|
|
|
|
// main formula: result = - sum_i(lables_i * log(softmax_i)) - sum over last dimension
|
|
// softmax_i = exp(logits_i) / sum_j(exp(logits_j))
|
|
// so result = sum_i( lables_i * (log(sum_j(exp(logits_j))) - logits_i) )
|
|
// for numerical stability we use shifted logits (one can approve this using simple math):
|
|
// softmax_i = exp(logits_i - maxLogit) / sum_j(exp(logits_j - maxLogit))
|
|
// maxLogit is max among logits_i
|
|
|
|
std::vector<LongType> dimensions = {-1};
|
|
NDArray* maxLogits = logits->reduceAlongDimension(reduce::Max, &dimensions, true);
|
|
NDArray* shiftedLogits_ptr = (*logits) - (*maxLogits);
|
|
delete maxLogits;
|
|
NDArray shiftedLogits = *shiftedLogits_ptr;
|
|
delete shiftedLogits_ptr;
|
|
|
|
NDArray* expShifted = shiftedLogits.transform(transform::Exp);
|
|
NDArray* sumExp = expShifted->reduceAlongDimension(reduce::Sum, &dimensions, true);
|
|
delete expShifted;
|
|
NDArray* logSumExp_ptr = sumExp->transform(transform::Log);
|
|
delete sumExp;
|
|
NDArray logSumExp = *logSumExp_ptr;
|
|
delete logSumExp_ptr;
|
|
|
|
// E = (newLabels * (logSumExp - shiftedLogits)).reduceAlongDimension(Sum)
|
|
NDArray* diff = logSumExp - shiftedLogits;
|
|
NDArray* product = (*newLabels) * (*diff);
|
|
delete diff;
|
|
NDArray* E_ptr = product->reduceAlongDimension(reduce::Sum, &dimensions);
|
|
delete product;
|
|
NDArray E = *E_ptr;
|
|
delete E_ptr;
|
|
|
|
// perform weights broadcasting/tile to E if it is necessary
|
|
auto weightsBroad = weights;
|
|
if (!weights->isScalar() && !weights->isSameShape(&E)) {
|
|
std::vector<LongType> weightsShape = {weights->lengthOf()};
|
|
if (E.rankOf() == 1 && weights->isVector() && weights->rankOf() > 1)
|
|
weightsBroad = weights->reshape(weights->ordering(), weightsShape);
|
|
else
|
|
weightsBroad = new NDArray(weights->tileToShape(E.shapeInfo()));
|
|
}
|
|
|
|
// multiply E on weights
|
|
E *= *weightsBroad;
|
|
|
|
switch (reductionMode) {
|
|
case 0: // 0 - "none", un-reduced weighted losses with the same shape as labels.
|
|
output->assign(&E);
|
|
break;
|
|
|
|
case 1: { // 1 - "weighted_sum", output is scalar and equal to sum of all elements of E array
|
|
E.reduceNumber(reduce::Sum, output);
|
|
break;
|
|
}
|
|
case 2: { // 2 - "weighted_mean", output is scalar and equal to sum of all elements of E array divided by sum of
|
|
// all elements of weightsBroad array
|
|
double sum;
|
|
if (weights->isScalar())
|
|
sum = weights->e<double>(0) * E.lengthOf();
|
|
else {
|
|
NDArray* sumPtr = weightsBroad->reduceNumber(reduce::Sum);
|
|
sum = sumPtr->e<double>(0);
|
|
delete sumPtr;
|
|
}
|
|
|
|
if (sum == 0.)
|
|
*output = 0.;
|
|
else {
|
|
NDArray* eSum = E.reduceNumber(reduce::Sum);
|
|
NDArray* result = (*eSum) / sum;
|
|
delete eSum;
|
|
output->assign(result);
|
|
delete result;
|
|
}
|
|
break;
|
|
}
|
|
case 3: { // 3 - "weighted_sum_by_nonzero_weights", output is scalar and equal to scalar sum of all elements of E
|
|
// array divided by number of non-zero weights
|
|
LongType numOfNonZeroWeights = 0;
|
|
if (weights->isScalar()) {
|
|
if (weights->e<double>(0) != 0.) numOfNonZeroWeights = E.lengthOf();
|
|
} else {
|
|
NDArray* countNonZero = weightsBroad->reduceNumber(reduce::CountNonZero);
|
|
numOfNonZeroWeights = countNonZero->e<LongType>(0);
|
|
delete countNonZero;
|
|
}
|
|
|
|
if (numOfNonZeroWeights == 0)
|
|
*output = 0.;
|
|
else {
|
|
NDArray* eSum = E.reduceNumber(reduce::Sum);
|
|
NDArray* result = (*eSum) / double(numOfNonZeroWeights);
|
|
delete eSum;
|
|
output->assign(result);
|
|
delete result;
|
|
}
|
|
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (weightsBroad != weights) delete weightsBroad;
|
|
|
|
if (newLabels != cLabels) delete newLabels;
|
|
|
|
delete cLabels;
|
|
|
|
return Status::OK;
|
|
}
|
|
|
|
//////////////////////////////////////////////////////////////////////////
|
|
DECLARE_TYPES(softmax_cross_entropy_loss) {
|
|
getOpDescriptor()
|
|
->setAllowedInputTypes(0, {ALL_FLOATS})
|
|
->setAllowedInputTypes(1, {ALL_FLOATS})
|
|
->setAllowedInputTypes(2, {ALL_FLOATS, ALL_INTS})
|
|
->setAllowedOutputTypes({ALL_FLOATS});
|
|
}
|
|
|
|
//////////////////////////////////////////////////////////////////////////
|
|
DECLARE_SHAPE_FN(softmax_cross_entropy_loss) {
|
|
auto logitsShapeInfo = inputShape->at(0);
|
|
auto weightsShapeInfo = inputShape->at(1);
|
|
auto labelsShapeInfo = inputShape->at(2);
|
|
|
|
// labels and logits must have the same shapes
|
|
REQUIRE_TRUE(shape::shapeEquals(logitsShapeInfo, labelsShapeInfo), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS OP: labels and logits arrays must have the same shapes, but got %s and %s "
|
|
"correspondingly!",
|
|
ShapeUtils::shapeAsString(labelsShapeInfo).c_str(), ShapeUtils::shapeAsString(logitsShapeInfo).c_str());
|
|
|
|
DataType outType = DataTypeUtils::pickFloatingType(ArrayOptions::dataType(logitsShapeInfo));
|
|
LongType* outShapeInfo = nullptr;
|
|
|
|
if (INT_ARG(0) != 0) // in this case output is scalar
|
|
outShapeInfo = ConstantShapeHelper::getInstance().scalarShapeInfo(outType);
|
|
else { // in this case output has the shape as labels and logits minus last dimension
|
|
std::vector<LongType> dimensions = {-1};
|
|
outShapeInfo = ShapeUtils::evalReduceShapeInfo(shape::order(logitsShapeInfo), &dimensions, logitsShapeInfo, false,
|
|
true, block.getWorkspace());
|
|
|
|
// weights array can be single scalar or has the same rank as output, and must be broadcastable to output
|
|
REQUIRE_TRUE(shape::isScalar(weightsShapeInfo) || shape::rank(weightsShapeInfo) == shape::rank(outShapeInfo), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS OP: weights array should be scalar or have the same rank as output array, "
|
|
"but got %i and %i correspondingly!",
|
|
shape::rank(weightsShapeInfo), shape::rank(outShapeInfo));
|
|
// check whether broadcast operation is possible for weights array
|
|
REQUIRE_TRUE(
|
|
shape::isScalar(weightsShapeInfo) || ShapeUtils::areShapesBroadcastable(weightsShapeInfo, outShapeInfo), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS OP: shapes of weights and output arrays should be broadcastable, but got weights = "
|
|
"%s and output = %s instead!",
|
|
ShapeUtils::shapeAsString(weightsShapeInfo).c_str(), ShapeUtils::shapeAsString(outShapeInfo).c_str());
|
|
}
|
|
|
|
return SHAPELIST(outShapeInfo);
|
|
}
|
|
|
|
//////////////////////////////////////////////////////////////////////////
|
|
CUSTOM_OP_IMPL(softmax_cross_entropy_loss_grad, 3, 3, false, 1, 1) {
|
|
auto logits = INPUT_VARIABLE(0);
|
|
auto weights = INPUT_VARIABLE(1);
|
|
auto labels = INPUT_VARIABLE(2);
|
|
|
|
auto dLdp = OUTPUT_VARIABLE(0); // dL/dlogits
|
|
auto dLdw = OUTPUT_VARIABLE(1); // dL/dweights
|
|
auto dLdl = OUTPUT_VARIABLE(2); // dL/dlabels
|
|
|
|
auto labelsSmoothing = T_ARG(0);
|
|
|
|
int reductionMode =
|
|
INT_ARG(0); // 0 - "none"; 1 - "weighted_sum"; 2 - "weighted_mean"; 3 - "weighted_sum_by_nonzero_weights"
|
|
// take into account Alex's proposition to treat "none" the same as "weighted_sum" mode when calculating gradients
|
|
if (reductionMode == 0) reductionMode = 1;
|
|
|
|
std::vector<LongType> *dimensions = new std::vector<LongType>({-1});
|
|
|
|
// input validation
|
|
REQUIRE_TRUE(labels->isSameShape(logits), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS_GRAD OP: labels and logits arrays must have the same shapes, but got %s and "
|
|
"%s correspondingly !",
|
|
ShapeUtils::shapeAsString(labels).c_str(), ShapeUtils::shapeAsString(logits).c_str());
|
|
// only 4 possible reduction modes exist
|
|
REQUIRE_TRUE(reductionMode == 0 || reductionMode == 1 || reductionMode == 2 || reductionMode == 3, 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS_GRAD OP: reduction mode value is not acceptable, possible values are 0, 1, "
|
|
"2, 3, but got %i instead!",
|
|
reductionMode);
|
|
auto lossShapeInfo = ShapeUtils::evalReduceShapeInfo(logits->ordering(), dimensions, logits->shapeInfo(), false,
|
|
false, block.getWorkspace());
|
|
// weights array can be single scalar or has the same shape as loss, and must be broadcastable to loss shape
|
|
REQUIRE_TRUE(weights->isScalar() || weights->rankOf() == shape::rank(lossShapeInfo), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS_GRAD OP: weights array should be scalar or have the same rank as loss "
|
|
"array, but got %i and %i correspondingly!",
|
|
weights->rankOf(), shape::rank(lossShapeInfo));
|
|
// check whether broadcast operation is possible for weights array
|
|
REQUIRE_TRUE(weights->isScalar() || ShapeUtils::areShapesBroadcastable(weights->shapeInfo(), lossShapeInfo), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS_GRAD OP: shapes of weights and loss arrays should be broadcastable, but got "
|
|
"weights = %s and loss = %s instead!",
|
|
ShapeUtils::shapeAsString(weights).c_str(), ShapeUtils::shapeAsString(lossShapeInfo).c_str());
|
|
// smoothing is possible for rank of logits/labels > 1
|
|
REQUIRE_TRUE(labels->rankOf() > 1 || (labels->rankOf() == 1 && labelsSmoothing == 0.), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS_GRAD OP: smoothing is not possible when rank of labels/ logits = 1 !");
|
|
|
|
// If label_smoothing is nonzero, smooth the labels towards 1/num_classes: new_onehot_labels = onehot_labels * (1 -
|
|
// label_smoothing) + label_smoothing / num_classes num_classes = labels->sizeAt(1)
|
|
NDArray* cLabels = labels->cast(weights->dataType()); // cast() already returns NDArray*
|
|
NDArray* newLabels = cLabels;
|
|
if (labelsSmoothing != 0.) {
|
|
newLabels = new NDArray(labels->shapeInfo(), dLdl->dataType(), false, block.launchContext());
|
|
NDArray* term1 = (1.f - labelsSmoothing) * (*cLabels);
|
|
NDArray* term2 = (*term1) + (labelsSmoothing / cLabels->sizeAt(1));
|
|
delete term1;
|
|
newLabels->assign(term2);
|
|
delete term2;
|
|
}
|
|
|
|
// Compute softmax
|
|
NDArray* maxLogits = logits->reduceAlongDimension(reduce::Max, dimensions, true);
|
|
NDArray* shiftedLogits_ptr = (*logits) - (*maxLogits);
|
|
delete maxLogits;
|
|
NDArray* expShifted = shiftedLogits_ptr->transform(transform::Exp);
|
|
delete shiftedLogits_ptr;
|
|
NDArray* sumExp = expShifted->reduceAlongDimension(reduce::Sum, dimensions, true);
|
|
NDArray* softmax_ptr = (*expShifted) / (*sumExp);
|
|
delete expShifted;
|
|
delete sumExp;
|
|
NDArray softmax = *softmax_ptr;
|
|
delete softmax_ptr;
|
|
|
|
// dEdp = softmax * sum_i(lables_i) - labels
|
|
NDArray* labelSum = newLabels->reduceAlongDimension(reduce::Sum, dimensions, true);
|
|
NDArray* softmaxTimesLabelSum = softmax * (*labelSum);
|
|
delete labelSum;
|
|
NDArray* dLdpTemp_ptr = (*softmaxTimesLabelSum) - (*newLabels);
|
|
delete softmaxTimesLabelSum;
|
|
dLdp->assign(dLdpTemp_ptr);
|
|
delete dLdpTemp_ptr;
|
|
|
|
// dEdl = -log(softmax)
|
|
NDArray* logSoftmax = softmax.transform(transform::Log);
|
|
NDArray negLogSoftmax = -(*logSoftmax); // unary negation returns value
|
|
delete logSoftmax;
|
|
NDArray* dLdlTemp_ptr = negLogSoftmax * (1.f - labelsSmoothing);
|
|
dLdl->assign(dLdlTemp_ptr);
|
|
delete dLdlTemp_ptr;
|
|
|
|
// Compute E for gradient calculations
|
|
NDArray* maxLogits2 = logits->reduceAlongDimension(reduce::Max, dimensions, true);
|
|
NDArray* shiftedLogits2_ptr = (*logits) - (*maxLogits2);
|
|
delete maxLogits2;
|
|
NDArray shiftedLogits = *shiftedLogits2_ptr;
|
|
delete shiftedLogits2_ptr;
|
|
|
|
NDArray* expShifted2 = shiftedLogits.transform(transform::Exp);
|
|
NDArray* sumExp2 = expShifted2->reduceAlongDimension(reduce::Sum, dimensions, true);
|
|
delete expShifted2;
|
|
NDArray* logSumExp_ptr = sumExp2->transform(transform::Log);
|
|
delete sumExp2;
|
|
NDArray logSumExp = *logSumExp_ptr;
|
|
delete logSumExp_ptr;
|
|
|
|
NDArray* diff = logSumExp - shiftedLogits;
|
|
NDArray* product = (*newLabels) * (*diff);
|
|
delete diff;
|
|
NDArray* E_ptr = product->reduceAlongDimension(reduce::Sum, dimensions);
|
|
delete product;
|
|
NDArray E = *E_ptr;
|
|
delete E_ptr;
|
|
|
|
// perform weights broadcasting/tile to E if it is necessary
|
|
auto weightsBroad = weights;
|
|
if (!weights->isScalar() && !weights->isSameShape(&E))
|
|
weightsBroad = new NDArray(weights->tileToShape(E.shapeInfo()));
|
|
|
|
auto excludeDims = ShapeUtils::evalDimsToExclude(dLdp->rankOf(), dimensions->size(), dimensions->data());
|
|
|
|
switch (reductionMode) {
|
|
case 1: { // 1 - "none" and "weighted_sum", output is scalar and equal to sum of all elements of E array
|
|
|
|
if (weights->isScalar() || weights->lengthOf() == 1) {
|
|
NDArray* eSum = E.reduceNumber(reduce::Sum);
|
|
dLdw->assign(eSum);
|
|
delete eSum;
|
|
*dLdp *= *weights;
|
|
*dLdl *= *weights;
|
|
} else {
|
|
dLdp->applyBroadcast(broadcast::Multiply, excludeDims, weightsBroad, dLdp);
|
|
dLdl->applyBroadcast(broadcast::Multiply, excludeDims, weightsBroad, dLdl);
|
|
|
|
if (weights != weightsBroad) {
|
|
std::vector<LongType> axesToReduceAlong =
|
|
ShapeUtils::evalBroadcastBackwardAxis(weights->shapeInfo(), weightsBroad->shapeInfo());
|
|
E.reduceAlongDimension(reduce::Sum, dLdw, &axesToReduceAlong, true);
|
|
} else
|
|
dLdw->assign(&E);
|
|
}
|
|
|
|
break;
|
|
}
|
|
case 2: { // 2 - "weighted_mean", output is scalar and equal to sum of all elements of E array divided by sum of
|
|
// all elements of weightsBroad array
|
|
NDArray* sum_ptr = nullptr;
|
|
if (weights->isScalar())
|
|
sum_ptr = (*weights) * E.lengthOf();
|
|
else
|
|
sum_ptr = weightsBroad->reduceNumber(reduce::Sum);
|
|
|
|
NDArray sum = *sum_ptr;
|
|
delete sum_ptr;
|
|
|
|
if (sum.e<double>(0) == 0.) {
|
|
*dLdp = 0.;
|
|
*dLdl = 0.;
|
|
*dLdw = 0.;
|
|
} else {
|
|
if (weights->isScalar() || weights->lengthOf() == 1) {
|
|
NDArray* temp_ptr = (*weights) / sum;
|
|
NDArray temp = *temp_ptr;
|
|
delete temp_ptr;
|
|
*dLdp *= temp;
|
|
*dLdl *= temp;
|
|
*dLdw = 0.;
|
|
} else {
|
|
NDArray* temp_ptr = (*weightsBroad) / sum;
|
|
NDArray temp = *temp_ptr;
|
|
delete temp_ptr;
|
|
dLdp->applyBroadcast(broadcast::Multiply, dimensions, &temp, dLdp);
|
|
dLdl->applyBroadcast(broadcast::Multiply, dimensions, &temp, dLdl);
|
|
|
|
if (weights != weightsBroad) {
|
|
std::vector<LongType> axesToReduceAlong =
|
|
ShapeUtils::evalBroadcastBackwardAxis(weights->shapeInfo(), weightsBroad->shapeInfo());
|
|
|
|
// Compute (E * sum - (E * weightsBroad).reduceNumber(Sum)) / (sum * sum)
|
|
NDArray* ETimesSum = E * sum;
|
|
NDArray* ETimesWeights = E * (*weightsBroad);
|
|
NDArray* ETimesWeightsSum = ETimesWeights->reduceNumber(reduce::Sum);
|
|
delete ETimesWeights;
|
|
|
|
NDArray* numerator = (*ETimesSum) - (*ETimesWeightsSum);
|
|
delete ETimesSum;
|
|
delete ETimesWeightsSum;
|
|
|
|
NDArray* sumSquared = sum * sum;
|
|
NDArray* result = (*numerator) / (*sumSquared);
|
|
delete numerator;
|
|
delete sumSquared;
|
|
|
|
result->reduceAlongDimension(reduce::Sum, dLdw, &axesToReduceAlong, true);
|
|
delete result;
|
|
} else {
|
|
// Compute (E * sum - (E * weightsBroad).reduceNumber(Sum)) / (sum * sum)
|
|
NDArray* ETimesSum = E * sum;
|
|
NDArray* ETimesWeights = E * (*weightsBroad);
|
|
NDArray* ETimesWeightsSum = ETimesWeights->reduceNumber(reduce::Sum);
|
|
delete ETimesWeights;
|
|
|
|
NDArray* numerator = (*ETimesSum) - (*ETimesWeightsSum);
|
|
delete ETimesSum;
|
|
delete ETimesWeightsSum;
|
|
|
|
NDArray* sumSquared = sum * sum;
|
|
NDArray* result = (*numerator) / (*sumSquared);
|
|
delete numerator;
|
|
delete sumSquared;
|
|
|
|
dLdw->assign(result);
|
|
delete result;
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
}
|
|
case 3: { // 3 - "weighted_sum_by_nonzero_weights", output is scalar and equal to scalar sum of all elements of E
|
|
// array divided by number of non-zero weights
|
|
LongType numOfNonZeroWeights = 0;
|
|
if (weights->isScalar()) {
|
|
if (weights->e<double>(0) != 0.) numOfNonZeroWeights = E.lengthOf();
|
|
} else {
|
|
NDArray* countNonZero = weightsBroad->reduceNumber(reduce::CountNonZero);
|
|
numOfNonZeroWeights = countNonZero->e<LongType>(0);
|
|
delete countNonZero;
|
|
}
|
|
|
|
if (numOfNonZeroWeights == 0) {
|
|
*dLdp = 0.;
|
|
*dLdl = 0.;
|
|
*dLdw = 0.;
|
|
} else {
|
|
if (weights->isScalar() || weights->lengthOf() == 1) {
|
|
NDArray* temp_ptr = (*weights) / numOfNonZeroWeights;
|
|
NDArray temp = *temp_ptr;
|
|
delete temp_ptr;
|
|
*dLdp *= temp;
|
|
*dLdl *= temp;
|
|
|
|
NDArray* eSum = E.reduceNumber(reduce::Sum);
|
|
NDArray* result = (*eSum) / numOfNonZeroWeights;
|
|
delete eSum;
|
|
dLdw->assign(result);
|
|
delete result;
|
|
} else {
|
|
NDArray* temp_ptr = (*weightsBroad) / numOfNonZeroWeights;
|
|
NDArray temp = *temp_ptr;
|
|
delete temp_ptr;
|
|
dLdp->applyBroadcast(broadcast::Multiply, dimensions, &temp, dLdp);
|
|
dLdl->applyBroadcast(broadcast::Multiply, dimensions, &temp, dLdl);
|
|
|
|
if (weights != weightsBroad) {
|
|
std::vector<LongType> axesToReduceAlong =
|
|
ShapeUtils::evalBroadcastBackwardAxis(weights->shapeInfo(), weightsBroad->shapeInfo());
|
|
E.reduceAlongDimension(reduce::Sum, dLdw, &axesToReduceAlong, true);
|
|
*dLdw /= numOfNonZeroWeights;
|
|
} else {
|
|
NDArray* eDivNum = E / numOfNonZeroWeights;
|
|
dLdw->assign(eDivNum);
|
|
delete eDivNum;
|
|
}
|
|
}
|
|
}
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (weightsBroad != weights) delete weightsBroad;
|
|
|
|
if (newLabels != cLabels) delete newLabels;
|
|
|
|
delete cLabels;
|
|
|
|
return Status::OK;
|
|
}
|
|
|
|
//////////////////////////////////////////////////////////////////////////
|
|
DECLARE_TYPES(softmax_cross_entropy_loss_grad) {
|
|
getOpDescriptor()
|
|
->setAllowedInputTypes(0, {ALL_FLOATS})
|
|
->setAllowedInputTypes(1, {ALL_FLOATS})
|
|
->setAllowedInputTypes(2, {ALL_FLOATS, ALL_INTS})
|
|
->setAllowedInputTypes(3, {ALL_FLOATS})
|
|
->setAllowedInputTypes(4, {ALL_FLOATS})
|
|
->setAllowedInputTypes(5, {ALL_FLOATS})
|
|
->setAllowedOutputTypes({ALL_FLOATS});
|
|
}
|
|
|
|
//////////////////////////////////////////////////////////////////////////
|
|
DECLARE_SHAPE_FN(softmax_cross_entropy_loss_grad) {
|
|
auto logitsShapeInfo = inputShape->at(0);
|
|
auto weightsShapeInfo = inputShape->at(1);
|
|
auto labelsShapeInfo = inputShape->at(2);
|
|
|
|
std::vector<LongType> dimensions = {-1};
|
|
|
|
// labels and logits must have the same shapes
|
|
REQUIRE_TRUE(shape::shapeEquals(logitsShapeInfo, labelsShapeInfo), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS_GRAD OP: labels and logits arrays must have the same shapes, but got %s and "
|
|
"%s correspondingly!",
|
|
ShapeUtils::shapeAsString(labelsShapeInfo).c_str(), ShapeUtils::shapeAsString(logitsShapeInfo).c_str());
|
|
auto lossShapeInfo = ShapeUtils::evalReduceShapeInfo(shape::order(logitsShapeInfo), &dimensions, logitsShapeInfo,
|
|
false, false, block.getWorkspace());
|
|
// weights array can be single scalar or has the same rank as loss, and must be broadcastable to loss
|
|
REQUIRE_TRUE(shape::isScalar(weightsShapeInfo) || shape::rank(weightsShapeInfo) == shape::rank(lossShapeInfo), 0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS_GRAD OP: weights array should be scalar or have the same rank as loss "
|
|
"array, but got %i and %i correspondingly!",
|
|
shape::rank(weightsShapeInfo), shape::rank(lossShapeInfo));
|
|
// check whether broadcast operation is possible for weights array
|
|
REQUIRE_TRUE(shape::isScalar(weightsShapeInfo) || ShapeUtils::areShapesBroadcastable(weightsShapeInfo, lossShapeInfo),
|
|
0,
|
|
"SOFTMAX_CROSS_ENTROPY_LOSS_GRAD OP: shapes of weights and loss arrays should be broadcastable, but got "
|
|
"weights = %s and loss = %s instead!",
|
|
ShapeUtils::shapeAsString(weightsShapeInfo).c_str(), ShapeUtils::shapeAsString(lossShapeInfo).c_str());
|
|
|
|
auto outType = DataTypeUtils::pickFloatingType(ArrayOptions::dataType(logitsShapeInfo));
|
|
|
|
auto dLdpShapeInfo = ConstantShapeHelper::getInstance().bufferForShapeInfo(outType, shape::order(logitsShapeInfo),
|
|
shape::rank(logitsShapeInfo),
|
|
shape::shapeOf(logitsShapeInfo))->primary();
|
|
|
|
auto dLdwShapeInfo = ConstantShapeHelper::getInstance().bufferForShapeInfo(outType, shape::order(weightsShapeInfo),
|
|
shape::rank(weightsShapeInfo),
|
|
shape::shapeOf(weightsShapeInfo))->primary();
|
|
|
|
auto dLdlShapeInfo = ConstantShapeHelper::getInstance().bufferForShapeInfo(outType, shape::order(labelsShapeInfo),
|
|
shape::rank(labelsShapeInfo),
|
|
shape::shapeOf(labelsShapeInfo))->primary();
|
|
return SHAPELIST(dLdpShapeInfo, dLdwShapeInfo, dLdlShapeInfo);
|
|
}
|
|
|
|
} // namespace ops
|
|
} // namespace sd
|
|
|
|
#endif
|