/******************************************************************************* * Copyright (c) 2015-2018 Skymind, Inc. * * 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. * * 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 ******************************************************************************/ // // Created by george@skymind.io on 6/1/2018. // @author Yurii Shyrma (iuriish@yahoo.com) // #include #include namespace nd4j { namespace ops { #if NOT_EXCLUDED(OP_reduce_dot_bp) //////////////////////////////////////////////////////////////////////////////// CUSTOM_OP_IMPL(reduce_dot_bp, 3, 2, false, 0, 0) { auto x = INPUT_VARIABLE(0); auto y = INPUT_VARIABLE(1); auto gradO = INPUT_VARIABLE(2); auto gradX = OUTPUT_VARIABLE(0); auto gradY = OUTPUT_VARIABLE(1); // L(x,y) = SUM(x_i * y_i) // dL/dx_i = y_i REQUIRE_TRUE(x->isSameShape(y), 0, "REDUCE_DOT_BP OP: both input arrays x and y should have same shapes, but got %s and %s correspondingly", ShapeUtils::shapeAsString(x).c_str(), ShapeUtils::shapeAsString(y).c_str()); if (gradO->lengthOf() == 1) { // scalar of reduced to scalar with keep dimensions gradX->assign((*y) * (*gradO)); gradY->assign((*x) * (*gradO)); } else { bool keepDims = false; auto dimensions = *block.getIArguments(); if (block.width() > 3) { auto axesVector = INPUT_VARIABLE(3); helpers::adjustAxis(x->rankOf(), axesVector, dimensions); } if (block.getBArguments()->size()) keepDims = B_ARG(0); else if (block.getTArguments()->size()) keepDims = (bool)T_ARG(0); REQUIRE_TRUE(dimensions.size() <= x->rankOf(), 0, "REDUCE_DOT_BP OP: the number of dimensions to reduce along must be <= input array rank, but got %i instead" , dimensions.size()); for(const auto& item : dimensions) REQUIRE_TRUE(item >= -x->rankOf() && item < x->rankOf(), 0, "REDUCE_DOT_BP OP: the input dimension to reduce along must be in range [-%i, %i), but got %i instead !" , x->rankOf(), x->rankOf(), item); if(!keepDims) { auto gradOShapeKeepDims = ShapeUtils::evalReduceShapeInfo(gradO->ordering(), dimensions, *x, true, false, block.getWorkspace()); auto r = gradO->reshape(gradO->ordering(), ShapeUtils::pullShapeFromShapeInfo(gradOShapeKeepDims)); // for example could be something like [a,b] -> [1,a,1,b] gradX->assign((*y) * r); gradY->assign((*x) * r); } else { gradX->assign((*y) * (*gradO)); gradY->assign((*x) * (*gradO)); } } return Status::OK(); } DECLARE_SHAPE_FN(reduce_dot_bp) { if(shape::length(inputShape->at(2)) > 1) { bool keepDims = false; auto dimensions = *block.getIArguments(); if (block.width() > 3) { auto axesVector = INPUT_VARIABLE(3); helpers::adjustAxis(INPUT_VARIABLE(0)->rankOf(), axesVector, dimensions); } if (block.getBArguments()->size()) keepDims = B_ARG(0); else if (block.getTArguments()->size()) keepDims = (bool)T_ARG(0); REQUIRE_TRUE(dimensions.size() <= inputShape->at(0)[0], 0, "REDUCE_DOT_BP OP: the number of dimensions to reduce along must be <= input array rank, but got %i instead" , dimensions.size()); for(const auto& item : dimensions) REQUIRE_TRUE(item >= -inputShape->at(0)[0] && item < inputShape->at(0)[0], 0, "REDUCE_DOT_BP OP: the input dimension to reduce along must be in range [-%i, %i), but got %i instead !" , inputShape->at(0)[0], inputShape->at(0)[0], item); } Nd4jLong *outShapeInfo1, *outShapeInfo2; COPY_SHAPE(inputShape->at(0), outShapeInfo1); COPY_SHAPE(inputShape->at(1), outShapeInfo2); return SHAPELIST(CONSTANT(outShapeInfo1), CONSTANT(outShapeInfo2)); } DECLARE_TYPES(reduce_dot_bp) { getOpDescriptor() ->setAllowedInputTypes(nd4j::DataType::ANY) ->setAllowedOutputTypes({ALL_FLOATS}); } #endif } }