org.deeplearning4j.nn.graph.vertex.impl.ReshapeVertex Maven / Gradle / Ivy

Go to download
/*
 *  ******************************************************************************
 *  *
 *  *
 *  * 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
 *  *****************************************************************************
 */

package org.deeplearning4j.nn.graph.vertex.impl;

import org.deeplearning4j.nn.api.Layer;
import org.deeplearning4j.nn.api.MaskState;
import org.deeplearning4j.nn.gradient.Gradient;
import org.deeplearning4j.nn.graph.ComputationGraph;
import org.deeplearning4j.nn.graph.vertex.BaseGraphVertex;
import org.deeplearning4j.nn.graph.vertex.VertexIndices;
import org.deeplearning4j.nn.workspace.ArrayType;
import org.deeplearning4j.nn.workspace.LayerWorkspaceMgr;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.common.primitives.Pair;

public class ReshapeVertex extends BaseGraphVertex {

    private char order;
    private int[] newShape;
    private int[] maskShape;


    public ReshapeVertex(ComputationGraph graph, String name, int vertexIndex, char order, int[] newShape, int[] maskShape, DataType dataType) {
        this(graph, name, vertexIndex, null, null, order, newShape, maskShape, dataType);
    }

    public ReshapeVertex(ComputationGraph graph, String name, int vertexIndex, VertexIndices[] inputVertices,
                    VertexIndices[] outputVertices, char order, int[] newShape, int[] maskShape, DataType dataType) {
        super(graph, name, vertexIndex, inputVertices, outputVertices, dataType);
        this.order = order;
        this.newShape = newShape;
        this.maskShape = maskShape;
    }

    @Override
    public boolean hasLayer() {
        return false;
    }

    @Override
    public Layer getLayer() {
        return null;
    }

    @Override
    public INDArray doForward(boolean training, LayerWorkspaceMgr workspaceMgr) {
        if (!canDoForward())
            throw new IllegalStateException("Cannot do forward pass: inputs not set");

        if (inputs.length > 1)
            throw new IllegalStateException("Reshape vertex requires a single input.");


        return workspaceMgr.dup(ArrayType.ACTIVATIONS, inputs[0].reshape(order, newShape));
    }

    @Override
    public Pair doBackward(boolean tbptt, LayerWorkspaceMgr workspaceMgr) {
        if (!canDoBackward())
            throw new IllegalStateException("Cannot do backward pass: errors not set");

        INDArray[] out = new INDArray[1];
        out[0] = workspaceMgr.dup(ArrayType.ACTIVATION_GRAD, epsilon.reshape(order, inputs[0].shape()));
        return new Pair<>(null, out);
    }

    @Override
    public void setBackpropGradientsViewArray(INDArray backpropGradientsViewArray) {
        if (backpropGradientsViewArray != null)
            throw new RuntimeException("Vertex does not have gradients; gradients view array cannot be set here");
    }

    @Override
    public Pair feedForwardMaskArrays(INDArray[] maskArrays, MaskState currentMaskState,
                    int minibatchSize) {
        if (maskArrays == null || maskArrays.length < 1 || maskArrays[0] == null) {
            return new Pair<>(null, currentMaskState);
        }

        if(maskShape != null){
            return new Pair<>(maskArrays[0].reshape(order, maskShape), currentMaskState);
        }

        //Mask array is an input mask. Therefore: 2 possible cases
        //(a) column vector mask (MLP, CNN), and
        //  i. output is rank 2 or 4 (MLP, CNN) -> no change
        // ii. output is rank 3 (RNN) -> to 2d
        //(b) 2d mask (RNN), and
        //  i. output is rank 2 or 4 (MLP, CNN) -> mask to column vector
        // ii. output is rank 3 (RNN) -> no change


        if(maskArrays[0].isColumnVectorOrScalar()){
            if(newShape.length == 2 || newShape.length == 4){
                return new Pair<>(maskArrays[0], currentMaskState);
            } else if(newShape.length == 3) {
                //Column vector -> 2d (FF -> RNN etc)
                int[] newMaskShape = new int[]{newShape[0], newShape[2]};
                return new Pair<>(maskArrays[0].reshape(order, newMaskShape), currentMaskState);
            }
        } else {
            if(newShape.length == 3){
                return new Pair<>(maskArrays[0], currentMaskState);
            } else {
                //RNN -> FF/CNN
                int[] newMaskShape = new int[]{newShape[0]*newShape[2], 1};
                return new Pair<>(maskArrays[0].reshape(order, newMaskShape), currentMaskState);
            }
        }

        //Other unknown case - shouldn't happen...
        return new Pair<>(maskArrays[0], currentMaskState);
    }

    @Override
    public String toString() {
        return "ReshapeVertex(id=" + this.getVertexIndex() + ",name=\"" + this.getVertexName() + "\",shape="
                        + newShape.toString() + ")";
    }
}