Convolution_tests.hpp

Go to the documentation of this file.

#ifndef CONVOLUTIONLAYERTESTS_HPP_
#define CONVOLUTIONLAYERTESTS_HPP_

#include <gtest/gtest.h>

// Redefine word "public" so every class field/method will be accessible for tests.
#define private public
#define protected public
#include <mlnn/convolution/Convolution.hpp>
#include <loss/LossTypes.hpp>

namespace mic { namespace neural_nets { namespace unit_tests {


class Conv2x2x2Filter2x1x1s1Double : public ::testing::Test {
public:
    // Constructor. Sets layer size.
    Conv2x2x2Filter2x1x1s1Double () : layer(2,2,2,2,1,1) {

        x = MAKE_MATRIX_PTR(double, 8, 1);

        desired_y = MAKE_MATRIX_PTR(double, 8, 1);

        dy = MAKE_MATRIX_PTR(double, 8, 1);

        target_y = MAKE_MATRIX_PTR(double, 8, 1);

        desired_dx = MAKE_MATRIX_PTR(double, 8, 1);
        // Number of filters * input channels.
        desired_dW = MAKE_MATRIX_PTR(double, 4, 1);
        // Number of "real neurons".
        desired_db = MAKE_MATRIX_PTR(double, 2, 1);
    }

protected:
    // Sets values
    virtual void SetUp() {
        (*layer.p["W0x0"]) << 0;
        (*layer.p["W0x1"]) << 2;

        (*layer.p["W1x0"]) << 3;
        (*layer.p["W1x1"]) << 1;

        // Set biases of both neurons.
        (*layer.p["b"]) << 0, 1;

        (*x).enumerate();// << 0, 1, 2, 3, 4, 5, 6, 7;
        (*desired_y) << 8, 10, 12, 14, 5, 9 ,13, 17;

        (*dy).enumerate();// << 0, 1, 2, 3, 4, 5, 6, 7;
        (*desired_dx) << 12, 15, 18, 21, 4, 7, 10, 13;
        (*target_y) << 13, 13, 20, 10, 3, 2, 12, 15;

        (*desired_dW) <<  14, 126, 38, 38;
        (*desired_db) <<  6, 22;
    }

private:
    mic::mlnn::convolution::Convolution<double> layer;

    // Loss function.
    mic::neural_nets::loss::SquaredErrorLoss<double> loss;

    mic::types::MatrixPtr<double> x;

    mic::types::MatrixPtr<double> desired_y;

    mic::types::MatrixPtr<double> dy;

    mic::types::MatrixPtr<double> target_y;

    mic::types::MatrixPtr<double> desired_dx;

    mic::types::MatrixPtr<double> desired_dW;

    mic::types::MatrixPtr<double> desired_db;

};



class Conv3x3x2Filter3x2x2s1Float : public ::testing::Test {
public:
    // Constructor. Sets layer size.
    Conv3x3x2Filter3x2x2s1Float () : layer(3,3,2,3,2,1) {

        x = MAKE_MATRIX_PTR(float, 18, 1);

        desired_y = MAKE_MATRIX_PTR(float, 12, 1);

    }

protected:
    // Sets values
    virtual void SetUp() {
        (*layer.p["W0x0"]) << 0, 1, 1, 0;
        (*layer.p["W0x1"]) << 0, -1, -1, 0;

        (*layer.p["W1x0"]) << -1, 0, 0, 1;
        (*layer.p["W1x1"]) << 1, 0, 0, -1;

        (*layer.p["W2x0"]) << 0, 0, 1, 1;
        (*layer.p["W2x1"]) << 0, 0, -1, -1;

        // Set biases of all three neurons.
        (*layer.p["b"]) << 1, 0, -1;

        (*x) << 1, 4, 7, 2, 5, 8, 3, 6, 9, 9, 6, 3, 8, 5, 2, 7, 4, 1;
        (*desired_y) << -7, 5, -3, 9, 8, 8, 8, 8, -7, 5, -3, 9;

    }

private:
    mic::mlnn::convolution::Convolution<float> layer;

    mic::types::MatrixPtr<float> x;

    mic::types::MatrixPtr<float> desired_y;

};


class Conv4x4x1Filter1x2x2s2Float : public ::testing::Test {
public:
    // Constructor. Sets layer size.
    Conv4x4x1Filter1x2x2s2Float () : layer(4,4,1,1,2,2) {

        x = MAKE_MATRIX_PTR(float, 16, 1);

        desired_y = MAKE_MATRIX_PTR(float, 4, 1);

        dy = MAKE_MATRIX_PTR(float, 4, 1);

        desired_dx = MAKE_MATRIX_PTR(float, 16, 1);

        desired_dW = MAKE_MATRIX_PTR(float, 4, 1);
        // Number of "real neurons".
        desired_db = MAKE_MATRIX_PTR(float, 1, 1);
    }

protected:
    // Sets values
    virtual void SetUp() {
        (*layer.p["W0x0"]) << 0, 1, 2, 3;

        // Set biases of both neurons.
        (*layer.p["b"]) << 0;

        (*x).enumerate();
        (*desired_y) << 24, 36, 72, 84;

        (*dy).enumerate();
        (*desired_dx) <<  0, 0, 0, 1, 0, 0, 2, 3, 0, 2, 0, 3, 4, 6, 6, 9;

        (*desired_dW) << 48, 54, 72, 78;
        (*desired_db) << 6;
    }

private:
    mic::mlnn::convolution::Convolution<float> layer;

    mic::types::MatrixPtr<float> x;

    mic::types::MatrixPtr<float> desired_y;

    mic::types::MatrixPtr<float> dy;

    mic::types::MatrixPtr<float> desired_dx;

    mic::types::MatrixPtr<float> desired_dW;

    mic::types::MatrixPtr<float> desired_db;

};


class Conv4x4x1Filter3x1x1s3Double : public ::testing::Test {
public:
    // Constructor. Sets layer size.
    Conv4x4x1Filter3x1x1s3Double () : layer(4,4,1,3,1,3) {

        x = MAKE_MATRIX_PTR(double, 16, 1);

        desired_y = MAKE_MATRIX_PTR(double, 12, 1);

        dy = MAKE_MATRIX_PTR(double, 12, 1);

        //target_y = MAKE_MATRIX_PTR(double, 12, 1);

        desired_dx = MAKE_MATRIX_PTR(double, 16, 1);
        // Number of filters * input channels.
        desired_dW = MAKE_MATRIX_PTR(double, 3, 1);
        // Number of "real neurons".
        desired_db = MAKE_MATRIX_PTR(double, 3, 1);
    }

protected:
    // Sets values
    virtual void SetUp() {
        (*layer.p["W0x0"]) << 0;
        (*layer.p["W1x0"]) << 1;
        (*layer.p["W2x0"]) << 2;

        // Set biases of neurons.
        (*layer.p["b"]) << -1, 0, 1;

        (*x).enumerate();
        (*desired_y) << -1, -1, -1, -1, 0, 3, 12,15, 1, 7, 25, 31;

        (*dy).enumerate();
        (*desired_dx) << 20, 0, 0, 23, 0,0,0,0,0,0,0,0, 26, 0,0, 29;
        //(*target_y) << 13, 13, 20, 10, 3, 2, 12, 15;

        (*desired_dW) <<  72, 192, 312;
        (*desired_db) <<  6, 22, 38;
    }

private:
    mic::mlnn::convolution::Convolution<double> layer;

    // Loss function.
    mic::neural_nets::loss::SquaredErrorLoss<double> loss;

    mic::types::MatrixPtr<double> x;

    mic::types::MatrixPtr<double> desired_y;

    mic::types::MatrixPtr<double> dy;

    //mic::types::MatrixPtr<double> target_y;

    mic::types::MatrixPtr<double> desired_dx;

    mic::types::MatrixPtr<double> desired_dW;

    mic::types::MatrixPtr<double> desired_db;

};

class Conv5x5x1Filter1x3x3s1Float : public ::testing::Test {
public:
    // Constructor. Sets layer size.
    Conv5x5x1Filter1x3x3s1Float () : layer(5,5,1,1,3,1) {

        x = MAKE_MATRIX_PTR(float, 25, 1);

        desired_y = MAKE_MATRIX_PTR(float, 9, 1);

        dy = MAKE_MATRIX_PTR(float, 9, 1);

    }

protected:
    // Sets values
    virtual void SetUp() {
        (*layer.p["W0x0"]) << 1, 0, 1, 0, 1, 0, 1, 0, 1;
        (*layer.p["b"]) << 0;

        (*x) << 1, 1, 1, 0, 0, 0, 1, 1, 1, 0, 0, 0, 1, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0;
        (*desired_y) << 4, 3, 4, 2, 4, 3, 2, 3, 4;

    }

private:
    mic::mlnn::convolution::Convolution<float> layer;

    mic::types::MatrixPtr<float> x;

    mic::types::MatrixPtr<float> desired_y;

    mic::types::MatrixPtr<float> dy;

};



class Conv5x5x1Filter1x2x2s3Float : public ::testing::Test {
public:
    // Constructor. Sets layer size.
    Conv5x5x1Filter1x2x2s3Float () : layer(5,5,1,1,2,3) {

        x = MAKE_MATRIX_PTR(float, 25, 1);

        desired_y = MAKE_MATRIX_PTR(float, 4, 1);

        dy = MAKE_MATRIX_PTR(float, 4, 1);

        desired_dx = MAKE_MATRIX_PTR(float, 25, 1);
        // Number of filters * input channels.
        desired_dW = MAKE_MATRIX_PTR(float, 4, 1);
        // Number of "real neurons".
        desired_db = MAKE_MATRIX_PTR(float, 1, 1);
    }

protected:
    // Sets values
    virtual void SetUp() {
        (*layer.p["W0x0"]).enumerate();
        (*layer.p["b"]) << 0;

        (*x).enumerate();
        (*desired_y) << 29, 47, 119, 137;

        (*dy).enumerate();
        (*desired_dx) << 0,0,0,0, 1, 0,0,0, 2, 3, 0,0,0,0,0,0, 2, 0,0, 3, 4, 6, 0, 6, 9;

        (*desired_dW) <<  87, 93, 117, 123;
        (*desired_db) <<  6;

    }

private:
    mic::mlnn::convolution::Convolution<float> layer;

    mic::types::MatrixPtr<float> x;

    mic::types::MatrixPtr<float> desired_y;

    mic::types::MatrixPtr<float> dy;

    mic::types::MatrixPtr<float> desired_dx;

    mic::types::MatrixPtr<float> desired_dW;

    mic::types::MatrixPtr<float> desired_db;
};


class Conv7x7x3Filter3x3x3s2Float : public ::testing::Test {
public:
    // Constructor. Sets layer size.
    Conv7x7x3Filter3x3x3s2Float () : layer(7,7,3,2,3,2) {

        x = MAKE_MATRIX_PTR(float, 7*7*3, 1);

        desired_y = MAKE_MATRIX_PTR(float, 3*3*2, 1);

    }

protected:
    // Sets values
    virtual void SetUp() {

        // Set weights of first neuron.
        (*layer.p["W0x0"]) << 0, -1, 0, 0, 1, -1, 1, 1, -1;
        (*layer.p["W0x1"]) << 1, 0, 1, 0, -1, -1, 1, 1, -1;
        (*layer.p["W0x2"]) << 1, 1, 0, -1, 1, -1, 1, 0, 1;

        // Set weights of second neuron.
        (*layer.p["W1x0"]) << 1, 1, -1, -1, -1, 1, 0, -1, -1;
        (*layer.p["W1x1"]) << 0, 1, 1, -1, 1, -1, 0, -1, -1;
        (*layer.p["W1x2"]) << 0, 0, 0, 1, 1, -1, -1, 0, 1;

        // Set biases of both neurons.
        (*layer.p["b"]) << 1, 0;

        (*x) <<
                // x[:,:,0]
                0, 0, 0, 0, 0, 0, 0,
                0, 0, 2, 1, 1, 2, 0,
                0, 2, 0, 1, 0, 1, 0,
                0, 2, 2, 2, 0, 0, 0,
                0, 2, 2, 1, 2, 0, 0,
                0, 0, 1, 1, 2, 0, 0,
                0, 0, 0, 0, 0, 0, 0,
                // x[:,:,1]
                0, 0, 0, 0, 0, 0, 0,
                0, 0, 2, 0, 2, 0, 0,
                0, 1, 2, 2, 2, 2, 0,
                0, 0, 0, 0, 1, 2, 0,
                0, 1, 0, 0, 1, 2, 0,
                0, 0, 0, 1, 0, 2, 0,
                0, 0, 0, 0, 0, 0, 0,
                // x[:,:,2]
                0, 0, 0, 0, 0, 0, 0,
                0, 0, 2, 1, 2, 2, 0,
                0, 1, 2, 0, 2, 0, 0,
                0, 2, 0, 1, 0, 2, 0,
                0, 1, 2, 2, 2, 0, 0,
                0, 2, 2, 0, 2, 0, 0,
                0, 0, 0, 0, 0, 0, 0;

        (*desired_y) <<
                // o[:,:,0]
                -2, 3, 10, 7, 12, 11, -1, -1, 0,
                // o[:,:,1]
                -5, -10, -6, 4, -3, 2, 2, 3, 6;
    }

private:
    mic::mlnn::convolution::Convolution<float> layer;

    mic::types::MatrixPtr<float> x;

    mic::types::MatrixPtr<float> desired_y;

};




class Conv5x6x1Filter1x4x4s1Float : public ::testing::Test {
public:
    // Constructor. Sets layer size.
    Conv5x6x1Filter1x4x4s1Float () : layer(5,6,1,1,4,1) {

        x = MAKE_MATRIX_PTR(float, 6, 5);

        desired_y = MAKE_MATRIX_PTR(float, 6,1);

        dy = MAKE_MATRIX_PTR(float, 6, 1);
        desired_dx = MAKE_MATRIX_PTR(float, 30, 1);
    }

protected:
    // Sets values
    virtual void SetUp() {
        for (size_t i=0; i<16; i++)
            (*layer.p["W0x0"])(i) = i+1;
        (*layer.p["W0x0"]).resize(4,4);
        (*layer.p["W0x0"]).transposeInPlace();
        //std::cout<<"*layer.p[W00] = \n" << (*layer.p["W00"]) << std::endl;
        (*layer.p["W0x0"]).resize(1, 4*4);

        // Set neuron bias.
        (*layer.p["b"]) << 0;

        for (size_t i=0; i<30; i++)
            (*x)(i) = i+1;
        (*x).transposeInPlace();
        //std::cout<<"*x = \n" << (*x) << std::endl;
        (*x).resize(30,1);

        (*desired_y) << 2064, 2880, 2200, 3016, 2336, 3152;
        //std::cout<<"*desired_y = \n" << (*desired_y) << std::endl;

        (*dy).enumerate();// << 1, 4, 2, 5, 3, 6;
        // IMPROPER VALUES!
        // (*desired_dx) <<  1, 9, 29, 49, 52, 4, 29, 77, 125, 121, 10, 62, 146, 230, 208, 16, 83, 167, 251, 223, 17, 75, 139, 203, 170, 12, 48, 84, 120, 96;
    }

private:
    mic::mlnn::convolution::Convolution<float> layer;

    mic::types::MatrixPtr<float> x;

    mic::types::MatrixPtr<float> desired_y;

    mic::types::MatrixPtr<float> dy;

    mic::types::MatrixPtr<float> desired_dx;

};


class Conv28x28x1Filter2x28x28s1Double : public ::testing::Test {
public:
    // Constructor. Sets layer size.
    Conv28x28x1Filter2x28x28s1Double () : layer(8,8,1,2,8,1) {

        x = MAKE_MATRIX_PTR(double, 8*8, 1);

        target_y = MAKE_MATRIX_PTR(double, 2, 1);

    }

protected:
    // Sets values
    virtual void SetUp() {

        // Random input
        (*x).randn(0, 6.0/(8*8));
        // We want output to be 1 and 0.
        (*target_y) << 1,0;

    }

private:
    mic::mlnn::convolution::Convolution<double> layer;

    // Loss function - cross entropy.
    mic::neural_nets::loss::SquaredErrorLoss<double> loss;

    mic::types::MatrixPtr<double> x;

    mic::types::MatrixPtr<double> target_y;

};


class Conv8x8x1Filter2x4x4s4Double : public ::testing::Test {
public:
    // Constructor. Sets layer size.
    Conv8x8x1Filter2x4x4s4Double () : layer(28,28,1,2,14,7) {

        x = MAKE_MATRIX_PTR(double, 28*28, 1);

        target_y = MAKE_MATRIX_PTR(double, 8, 1);

    }

protected:
    // Sets values
    virtual void SetUp() {

        // Random input
        (*x).randn(0, 6.0/(8*8));
        // We want output to be 1 and 0.
        //(*target_y) << 0.25, 0.5, 1, 0, 0.2, 0.1, 0.3, 0.5;

    }

private:
    mic::mlnn::convolution::Convolution<double> layer;

    // Loss function - cross entropy.
    mic::neural_nets::loss::SquaredErrorLoss<double> loss;

    mic::types::MatrixPtr<double> x;

    mic::types::MatrixPtr<double> target_y;

};


} } } //: namespaces

#endif /* CONVOLUTIONLAYERTESTS_HPP_ */