• Home
• org.nd4j
• nd4j-api
• 1.0.0-beta7
• source code
• CheckUtil.java

×

Project download

Many resources are needed to download a project. Please understand that we have to compensate our server costs. Thank you in advance.
Project price only 1 $

You can buy this project and download/modify it how often you want.

org.nd4j.linalg.checkutil.CheckUtil Maven / Gradle / Ivy

/*******************************************************************************
 * 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
 ******************************************************************************/

package org.nd4j.linalg.checkutil;

import lombok.val;
import org.apache.commons.math3.linear.BlockRealMatrix;
import org.apache.commons.math3.linear.RealMatrix;
import org.nd4j.linalg.api.buffer.DataType;
import org.nd4j.linalg.api.ndarray.INDArray;
import org.nd4j.linalg.api.shape.Shape;
import org.nd4j.linalg.exception.ND4JArraySizeException;
import org.nd4j.linalg.factory.Nd4j;

import java.util.Arrays;

/**@author Alex Black
 */
public class CheckUtil {

    /**Check first.mmul(second) using Apache commons math mmul. Float/double matrices only.

     * Returns true if OK, false otherwise.

     * Checks each element according to relative error (|a-b|/(|a|+|b|); however absolute error |a-b| must
     * also exceed minAbsDifference for it to be considered a failure. This is necessary to avoid instability
     * near 0: i.e., Nd4j mmul might return element of 0.0 (due to underflow on float) while Apache commons math
     * mmul might be say 1e-30 or something (using doubles). 
     * Throws exception if matrices can't be multiplied
     * Checks each element of the result. If
     * @param first First matrix
     * @param second Second matrix
     * @param maxRelativeDifference Maximum relative error
     * @param minAbsDifference Minimum absolute difference for failure
     * @return true if OK, false if result incorrect
     */
    public static boolean checkMmul(INDArray first, INDArray second, double maxRelativeDifference,
                    double minAbsDifference) {
        if (first.size(1) != second.size(0))
            throw new IllegalArgumentException("first.columns != second.rows");
        RealMatrix rmFirst = convertToApacheMatrix(first);
        RealMatrix rmSecond = convertToApacheMatrix(second);

        INDArray result = first.mmul(second);
        RealMatrix rmResult = rmFirst.multiply(rmSecond);

        if (!checkShape(rmResult, result))
            return false;
        boolean ok = checkEntries(rmResult, result, maxRelativeDifference, minAbsDifference);
        if (!ok) {
            INDArray onCopies = Shape.toOffsetZeroCopy(first).mmul(Shape.toOffsetZeroCopy(second));
            printFailureDetails(first, second, rmResult, result, onCopies, "mmul");
        }
        return ok;
    }

    public static boolean checkGemm(INDArray a, INDArray b, INDArray c, boolean transposeA, boolean transposeB,
                    double alpha, double beta, double maxRelativeDifference, double minAbsDifference) {
        long commonDimA = (transposeA ? a.rows() : a.columns());
        long commonDimB = (transposeB ? b.columns() : b.rows());
        if (commonDimA != commonDimB)
            throw new IllegalArgumentException("Common dimensions don't match: a.shape=" + Arrays.toString(a.shape())
                            + ", b.shape=" + Arrays.toString(b.shape()) + ", tA=" + transposeA + ", tb=" + transposeB);
        long outRows = (transposeA ? a.columns() : a.rows());
        long outCols = (transposeB ? b.rows() : b.columns());
        if (c.rows() != outRows || c.columns() != outCols)
            throw new IllegalArgumentException("C does not match outRows or outCols");
        if (c.offset() != 0 || c.ordering() != 'f')
            throw new IllegalArgumentException("Invalid c");

        INDArray aConvert = transposeA ? a.transpose() : a;
        RealMatrix rmA = convertToApacheMatrix(aConvert);
        INDArray bConvet = transposeB ? b.transpose() : b;
        RealMatrix rmB = convertToApacheMatrix(bConvet);
        RealMatrix rmC = convertToApacheMatrix(c);
        RealMatrix rmExpected = rmA.scalarMultiply(alpha).multiply(rmB).add(rmC.scalarMultiply(beta));
        INDArray cCopy1 = Nd4j.create(c.shape(), 'f');
        cCopy1.assign(c);
        INDArray cCopy2 = Nd4j.create(c.shape(), 'f');
        cCopy2.assign(c);

        INDArray out = Nd4j.gemm(a, b, c, transposeA, transposeB, alpha, beta);
        if (out != c) {
            System.out.println("Returned different array than c");
            return false;
        }
        if (!checkShape(rmExpected, out))
            return false;
        boolean ok = checkEntries(rmExpected, out, maxRelativeDifference, minAbsDifference);
        if (!ok) {
            INDArray aCopy = Shape.toOffsetZeroCopy(a);
            INDArray bCopy = Shape.toOffsetZeroCopy(b);
            INDArray onCopies = Nd4j.gemm(aCopy, bCopy, cCopy1, transposeA, transposeB, alpha, beta);
            printGemmFailureDetails(a, b, cCopy2, transposeA, transposeB, alpha, beta, rmExpected, out, onCopies);
        }
        return ok;
    }

    /**Same as checkMmul, but for matrix addition */
    public static boolean checkAdd(INDArray first, INDArray second, double maxRelativeDifference,
                    double minAbsDifference) {
        RealMatrix rmFirst = convertToApacheMatrix(first);
        RealMatrix rmSecond = convertToApacheMatrix(second);

        INDArray result = first.add(second);
        RealMatrix rmResult = rmFirst.add(rmSecond);

        if (!checkShape(rmResult, result))
            return false;
        boolean ok = checkEntries(rmResult, result, maxRelativeDifference, minAbsDifference);
        if (!ok) {
            INDArray onCopies = Shape.toOffsetZeroCopy(first).add(Shape.toOffsetZeroCopy(second));
            printFailureDetails(first, second, rmResult, result, onCopies, "add");
        }
        return ok;
    }

    /** Same as checkMmul, but for matrix subtraction */
    public static boolean checkSubtract(INDArray first, INDArray second, double maxRelativeDifference,
                    double minAbsDifference) {
        RealMatrix rmFirst = convertToApacheMatrix(first);
        RealMatrix rmSecond = convertToApacheMatrix(second);

        INDArray result = first.sub(second);
        RealMatrix rmResult = rmFirst.subtract(rmSecond);

        if (!checkShape(rmResult, result))
            return false;
        boolean ok = checkEntries(rmResult, result, maxRelativeDifference, minAbsDifference);
        if (!ok) {
            INDArray onCopies = Shape.toOffsetZeroCopy(first).sub(Shape.toOffsetZeroCopy(second));
            printFailureDetails(first, second, rmResult, result, onCopies, "sub");
        }
        return ok;
    }

    public static boolean checkMulManually(INDArray first, INDArray second, double maxRelativeDifference,
                    double minAbsDifference) {
        //No apache commons element-wise multiply, but can do this manually

        INDArray result = first.mul(second);
        long[] shape = first.shape();

        INDArray expected = Nd4j.zeros(first.shape());

        for (int i = 0; i < shape[0]; i++) {
            for (int j = 0; j < shape[1]; j++) {
                double v = first.getDouble(i, j) * second.getDouble(i, j);
                expected.putScalar(new int[] {i, j}, v);
            }
        }
        if (!checkShape(expected, result))
            return false;
        boolean ok = checkEntries(expected, result, maxRelativeDifference, minAbsDifference);
        if (!ok) {
            INDArray onCopies = Shape.toOffsetZeroCopy(first).mul(Shape.toOffsetZeroCopy(second));
            printFailureDetails(first, second, expected, result, onCopies, "mul");
        }
        return ok;
    }

    public static boolean checkDivManually(INDArray first, INDArray second, double maxRelativeDifference,
                    double minAbsDifference) {
        //No apache commons element-wise division, but can do this manually

        INDArray result = first.div(second);
        long[] shape = first.shape();

        INDArray expected = Nd4j.zeros(first.shape());

        for (int i = 0; i < shape[0]; i++) {
            for (int j = 0; j < shape[1]; j++) {
                double v = first.getDouble(i, j) / second.getDouble(i, j);
                expected.putScalar(new int[] {i, j}, v);
            }
        }
        if (!checkShape(expected, result))
            return false;
        boolean ok = checkEntries(expected, result, maxRelativeDifference, minAbsDifference);
        if (!ok) {
            INDArray onCopies = Shape.toOffsetZeroCopy(first).mul(Shape.toOffsetZeroCopy(second));
            printFailureDetails(first, second, expected, result, onCopies, "div");
        }
        return ok;
    }

    private static boolean checkShape(RealMatrix rmResult, INDArray result) {
        long[] outShape = {rmResult.getRowDimension(), rmResult.getColumnDimension()};
        if (!Arrays.equals(outShape, result.shape())) {
            System.out.println("Failure on shape: " + Arrays.toString(result.shape()) + ", expected "
                            + Arrays.toString(outShape));
            return false;
        }
        return true;
    }

    private static boolean checkShape(INDArray expected, INDArray actual) {
        if (!Arrays.equals(expected.shape(), actual.shape())) {
            System.out.println("Failure on shape: " + Arrays.toString(actual.shape()) + ", expected "
                            + Arrays.toString(expected.shape()));
            return false;
        }
        return true;
    }

    public static boolean checkEntries(RealMatrix rmResult, INDArray result, double maxRelativeDifference,
                    double minAbsDifference) {
        int[] outShape = {rmResult.getRowDimension(), rmResult.getColumnDimension()};
        for (int i = 0; i < outShape[0]; i++) {
            for (int j = 0; j < outShape[1]; j++) {
                double expOut = rmResult.getEntry(i, j);
                double actOut = result.getDouble(i, j);

                if (Double.isNaN(actOut)) {
                    System.out.println("NaN failure on value: (" + i + "," + j + " exp=" + expOut + ", act=" + actOut);
                    return false;
                }

                if (expOut == 0.0 && actOut == 0.0)
                    continue;
                double absError = Math.abs(expOut - actOut);
                double relError = absError / (Math.abs(expOut) + Math.abs(actOut));
                if (relError > maxRelativeDifference && absError > minAbsDifference) {
                    System.out.println("Failure on value: (" + i + "," + j + " exp=" + expOut + ", act=" + actOut
                                    + ", absError=" + absError + ", relError=" + relError);
                    return false;
                }
            }
        }
        return true;
    }

    public static boolean checkEntries(INDArray expected, INDArray actual, double maxRelativeDifference,
                    double minAbsDifference) {
        long[] outShape = expected.shape();
        for (int i = 0; i < outShape[0]; i++) {
            for (int j = 0; j < outShape[1]; j++) {
                double expOut = expected.getDouble(i, j);
                double actOut = actual.getDouble(i, j);
                if (expOut == 0.0 && actOut == 0.0)
                    continue;
                double absError = Math.abs(expOut - actOut);
                double relError = absError / (Math.abs(expOut) + Math.abs(actOut));
                if (relError > maxRelativeDifference && absError > minAbsDifference) {
                    System.out.println("Failure on value: (" + i + "," + j + " exp=" + expOut + ", act=" + actOut
                                    + ", absError=" + absError + ", relError=" + relError);
                    return false;
                }
            }
        }
        return true;
    }

    public static RealMatrix convertToApacheMatrix(INDArray matrix) {
        if (matrix.rank() != 2)
            throw new IllegalArgumentException("Input rank is not 2 (not matrix)");
        long[] shape = matrix.shape();

        if (matrix.columns() > Integer.MAX_VALUE || matrix.rows() > Integer.MAX_VALUE)
            throw new ND4JArraySizeException();

        BlockRealMatrix out = new BlockRealMatrix((int) shape[0], (int) shape[1]);
        for (int i = 0; i < shape[0]; i++) {
            for (int j = 0; j < shape[1]; j++) {
                double value = matrix.getDouble(i, j);
                out.setEntry(i, j, value);
            }
        }
        return out;
    }

    public static INDArray convertFromApacheMatrix(RealMatrix matrix, DataType dataType) {
        val shape = new long[] {matrix.getRowDimension(), matrix.getColumnDimension()};
        INDArray out = Nd4j.create(dataType, shape);
        for (int i = 0; i < shape[0]; i++) {
            for (int j = 0; j < shape[1]; j++) {
                double value = matrix.getEntry(i, j);
                out.putScalar(new int[] {i, j}, value);
            }
        }
        return out;
    }



    public static void printFailureDetails(INDArray first, INDArray second, RealMatrix expected, INDArray actual,
                    INDArray onCopies, String op) {
        System.out.println("\nFactory: " + Nd4j.factory().getClass() + "\n");

        System.out.println("First:");
        printMatrixFullPrecision(first);
        System.out.println("\nSecond:");
        printMatrixFullPrecision(second);
        System.out.println("\nExpected (Apache Commons)");
        printApacheMatrix(expected);
        System.out.println("\nSame Nd4j op on copies: (Shape.toOffsetZeroCopy(first)." + op
                        + "(Shape.toOffsetZeroCopy(second)))");
        printMatrixFullPrecision(onCopies);
        System.out.println("\nActual:");
        printMatrixFullPrecision(actual);
    }

    public static void printGemmFailureDetails(INDArray a, INDArray b, INDArray c, boolean transposeA,
                    boolean transposeB, double alpha, double beta, RealMatrix expected, INDArray actual,
                    INDArray onCopies) {
        System.out.println("\nFactory: " + Nd4j.factory().getClass() + "\n");
        System.out.println("Op: gemm(a,b,c,transposeA=" + transposeA + ",transposeB=" + transposeB + ",alpha=" + alpha
                        + ",beta=" + beta + ")");

        System.out.println("a:");
        printMatrixFullPrecision(a);
        System.out.println("\nb:");
        printMatrixFullPrecision(b);
        System.out.println("\nc:");
        printMatrixFullPrecision(c);
        System.out.println("\nExpected (Apache Commons)");
        printApacheMatrix(expected);
        System.out.println("\nSame Nd4j op on zero offset copies: gemm(aCopy,bCopy,cCopy," + transposeA + ","
                        + transposeB + "," + alpha + "," + beta + ")");
        printMatrixFullPrecision(onCopies);
        System.out.println("\nActual:");
        printMatrixFullPrecision(actual);
    }

    public static void printMatrixFullPrecision(INDArray matrix) {
        boolean floatType = (matrix.data().dataType() == DataType.FLOAT);
        printNDArrayHeader(matrix);
        long[] shape = matrix.shape();
        for (int i = 0; i < shape[0]; i++) {
            for (int j = 0; j < shape[1]; j++) {
                if (floatType)
                    System.out.print(matrix.getFloat(i, j));
                else
                    System.out.print(matrix.getDouble(i, j));
                if (j != shape[1] - 1)
                    System.out.print(", ");
                else
                    System.out.println();
            }
        }
    }

    public static void printNDArrayHeader(INDArray array) {
        System.out.println(array.data().dataType() + " - order=" + array.ordering() + ", offset=" + array.offset()
                        + ", shape=" + Arrays.toString(array.shape()) + ", stride=" + Arrays.toString(array.stride())
                        + ", length=" + array.length() + ", data().length()=" + array.data().length());
    }

    public static void printFailureDetails(INDArray first, INDArray second, INDArray expected, INDArray actual,
                    INDArray onCopies, String op) {
        System.out.println("\nFactory: " + Nd4j.factory().getClass() + "\n");

        System.out.println("First:");
        printMatrixFullPrecision(first);
        System.out.println("\nSecond:");
        printMatrixFullPrecision(second);
        System.out.println("\nExpected");
        printMatrixFullPrecision(expected);
        System.out.println("\nSame Nd4j op on copies: (Shape.toOffsetZeroCopy(first)." + op
                        + "(Shape.toOffsetZeroCopy(second)))");
        printMatrixFullPrecision(onCopies);
        System.out.println("\nActual:");
        printMatrixFullPrecision(actual);
    }

    public static void printApacheMatrix(RealMatrix matrix) {
        int nRows = matrix.getRowDimension();
        int nCols = matrix.getColumnDimension();
        System.out.println("Apache Commons RealMatrix: Shape: [" + nRows + "," + nCols + "]");
        for (int i = 0; i < nRows; i++) {
            for (int j = 0; j < nCols; j++) {
                System.out.print(matrix.getEntry(i, j));
                if (j != nCols - 1)
                    System.out.print(", ");
                else
                    System.out.println();
            }
        }
    }
}