org.nd4j.linalg.api.ops.impl.reduce.bp.BaseReductionBp Maven / Gradle / Ivy

Show more of this group Show more artifacts with this name
Show all versions of nd4j-api Show documentation
The newest version!
/*
 *  ******************************************************************************
 *  *
 *  *
 *  * 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.nd4j.linalg.api.ops.impl.reduce.bp;

import lombok.NoArgsConstructor;
import org.nd4j.autodiff.samediff.SDVariable;
import org.nd4j.autodiff.samediff.SameDiff;
import org.nd4j.common.base.Preconditions;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.ops.DynamicCustomOp;

import java.util.Collections;
import java.util.List;


@NoArgsConstructor
public abstract class BaseReductionBp extends DynamicCustomOp {

    protected boolean keepDims;
    protected int[] dimensions;

    /**
     *
     * @param origInput    Pre-reduced input
     * @param gradAtOutput Gradient at the output
     * @param keepDims     If true: reduction dimensions were kept
     * @param dimensions   Dimensions to reduce. May be null
     */
    public BaseReductionBp(SameDiff sameDiff, SDVariable origInput, SDVariable gradAtOutput, boolean keepDims, int... dimensions) {
        super(null, sameDiff, new SDVariable[]{origInput, gradAtOutput}, false);
        this.keepDims = keepDims;
        this.dimensions = dimensions;
    }

    /**
     *
     * @param origInput1   Pre-reduced input 1
     * @param origInput2   Pre-reduced input 2
     * @param gradAtOutput Gradient at the output
     * @param keepDims     If true: reduction dimensions were kept
     * @param dimensions   Dimensions to reduce. May be null
     */
    public BaseReductionBp(SameDiff sameDiff, SDVariable origInput1, SDVariable origInput2, SDVariable gradAtOutput, boolean keepDims, int... dimensions) {
        super(null, sameDiff, new SDVariable[]{origInput1, origInput2, gradAtOutput}, false);
        this.keepDims = keepDims;
        this.dimensions = dimensions;
    }

    /**
     *
     * @param origInput    Pre-reduced input
     * @param gradAtOutput Gradient at the output
     * @param output       Output array - i.e., gradient at the input to the reduction function
     * @param keepDims     If true: reduction dimensions were kept
     * @param dimensions   Dimensions to reduce. May be null
     */
    public BaseReductionBp(INDArray origInput, INDArray gradAtOutput, INDArray output, boolean keepDims, int... dimensions) {
        super(null, new INDArray[]{origInput, gradAtOutput}, (output == null ? null : new INDArray[]{output}));
        this.keepDims = keepDims;
        this.dimensions = dimensions;
    }

    /**
     *
     * @param origInput1   Pre-reduced input1
     * @param origInput2   Pre-reduced input2
     * @param gradAtOutput Gradient at the output
     * @param output       Output array - i.e., gradient at the input to the reduction function
     * @param keepDims     If true: reduction dimensions were kept
     * @param dimensions   Dimensions to reduce. May be null
     */
    public BaseReductionBp(INDArray origInput1, INDArray origInput2, INDArray gradAtOutput, INDArray output, boolean keepDims, int... dimensions){
        super(null, new INDArray[]{origInput1, origInput2, gradAtOutput}, (output == null ? null : new INDArray[]{output}));
        this.keepDims = keepDims;
        this.dimensions = dimensions;
    }

    public BaseReductionBp(INDArray origInput1, INDArray origInput2, INDArray gradAtOutput, INDArray output1, INDArray output2, boolean keepDims, int... dimensions){
        super(null, new INDArray[]{origInput1, origInput2, gradAtOutput}, new INDArray[]{output1, output2});
        this.keepDims = keepDims;
        this.dimensions = dimensions;
    }

    public BaseReductionBp(INDArray origInput, INDArray gradAtOutput, INDArray output, boolean keepDims, INDArray dimensions) {
        super(null,new INDArray[]{origInput,gradAtOutput,dimensions},new INDArray[]{output});
        this.keepDims = keepDims;
    }

    public BaseReductionBp(SameDiff sameDiff, SDVariable origInput, SDVariable gradAtOutput, boolean keepDims, SDVariable dimensions) {
        super(null,sameDiff,new SDVariable[]{origInput,gradAtOutput,dimensions},false);
        this.keepDims = keepDims;
    }

    protected void addArgs() {
        addBArgument(keepDims);
        if(dimensions != null && dimensions.length > 0) {
            if(dimensions.length != 1 || dimensions[0] != Integer.MAX_VALUE) {
                //Integer.MAX_VALUE means "full array" but here no dimension args == full array
                addIArgument(dimensions);
            }
        }
    }

    public abstract String opName();


    @Override
    public List calculateOutputDataTypes(List dataTypes) {
        //Reduction backprop ops: expect 2 inputs... the original input, and the gradient at the outputs
        //For example, for y=mean(x), inputs to ReduceMeanBp are x and dL/dy; output is dL/dx
        //Now, we expect gradient dL/dx datatype to be same as x - which resticts us to real-valued x input
        //i.e., 'gradient' of integer or boolean isn't defined
        Preconditions.checkState(dataTypes != null && dataTypes.size() == 2, "Expected exactly 2 input datatype for %s, got input %s", getClass(), dataTypes);
        Preconditions.checkState(dataTypes.get(0).isFPType(), "First input must be a floating point type, got %s", dataTypes.get(0));
        Preconditions.checkState(dataTypes.get(1).isFPType(), "Second input (gradient at reduction output) must be a floating point type, got %s", dataTypes.get(1));
        return Collections.singletonList(dataTypes.get(0));
    }

}