Mathis/CUDANet
1
0
Fork 0
mirror of https://github.com/lordmathis/CUDANet.git synced 2025-11-06 01:34:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

Abstract activation and implement softmax

Browse Source
This commit is contained in:
Matúš Námešný
2024-03-17 18:37:15 +01:00
parent b1621819ca
commit 42d646750b
19 changed files with 370 additions and 205 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
test/CMakeLists.txt
View File

@@ -2,10 +2,11 @@ find_package(GTest REQUIRED)
include_directories(${GTEST_INCLUDE_DIRS})
add_executable(test_main
EXCLUDE_FROM_ALL
layers/test_dense.cu
layers/test_conv2d.cu
layers/test_input.cu
kernels/test_activations.cu
kernels/test_activation_functions.cu
kernels/test_padding.cu
kernels/test_matmul.cu
)

2
test/kernels/test_activations.cu → test/kernels/test_activation_functions.cu
View File

@@ -3,7 +3,7 @@
#include <iostream>
#include "activations.cuh"
#include "activation_functions.cuh"
TEST(ActivationsTest, SigmoidSanityCheck) {

81
test/layers/test_conv2d.cu
View File

@@ -8,21 +8,21 @@
class Conv2dTest : public ::testing::Test {
protected:
CUDANet::Layers::Conv2d commonTestSetup(
int inputSize,
int inputChannels,
int kernelSize,
int stride,
CUDANet::Layers::Padding padding,
int numFilters,
CUDANet::Layers::Activation activation,
std::vector<float>& input,
float* kernels,
float*& d_input
int inputSize,
int inputChannels,
int kernelSize,
int stride,
int numFilters,
CUDANet::Layers::Padding padding,
CUDANet::Layers::ActivationType activationType,
std::vector<float>& input,
float* kernels,
float*& d_input
) {
// Create Conv2d layer
CUDANet::Layers::Conv2d conv2d(
inputSize, inputChannels, kernelSize, stride, padding, numFilters,
activation
inputSize, inputChannels, kernelSize, stride, numFilters, padding,
activationType
);
conv2d.setWeights(kernels);
@@ -53,13 +53,14 @@ class Conv2dTest : public ::testing::Test {
};
TEST_F(Conv2dTest, SimpleTest) {
int inputSize = 4;
int inputChannels = 1;
int kernelSize = 2;
int stride = 1;
CUDANet::Layers::Padding padding = CUDANet::Layers::Padding::VALID;
int numFilters = 1;
CUDANet::Layers::Activation activation = CUDANet::Layers::Activation::NONE;
int inputSize = 4;
int inputChannels = 1;
int kernelSize = 2;
int stride = 1;
int numFilters = 1;
CUDANet::Layers::Padding padding = CUDANet::Layers::Padding::VALID;
CUDANet::Layers::ActivationType activationType =
CUDANet::Layers::ActivationType::NONE;
std::vector<float> input = {1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f,
7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f,
@@ -75,8 +76,8 @@ TEST_F(Conv2dTest, SimpleTest) {
float* d_output;
CUDANet::Layers::Conv2d conv2d = commonTestSetup(
inputSize, inputChannels, kernelSize, stride, padding, numFilters,
activation, input, kernels.data(), d_input
inputSize, inputChannels, kernelSize, stride, numFilters, padding,
activationType, input, kernels.data(), d_input
);
int outputSize = (inputSize - kernelSize) / stride + 1;
@@ -102,13 +103,14 @@ TEST_F(Conv2dTest, SimpleTest) {
}
TEST_F(Conv2dTest, PaddedTest) {
int inputSize = 5;
int inputChannels = 3;
int kernelSize = 3;
int stride = 1;
CUDANet::Layers::Padding padding = CUDANet::Layers::Padding::SAME;
int numFilters = 2;
CUDANet::Layers::Activation activation = CUDANet::Layers::Activation::NONE;
int inputSize = 5;
int inputChannels = 3;
int kernelSize = 3;
int stride = 1;
int numFilters = 2;
CUDANet::Layers::Padding padding = CUDANet::Layers::Padding::SAME;
CUDANet::Layers::ActivationType activationType =
CUDANet::Layers::ActivationType::NONE;
// clang-format off
std::vector<float> input = {
@@ -164,8 +166,8 @@ TEST_F(Conv2dTest, PaddedTest) {
float* d_output;
CUDANet::Layers::Conv2d conv2d = commonTestSetup(
inputSize, inputChannels, kernelSize, stride, padding, numFilters,
activation, input, kernels.data(), d_input
inputSize, inputChannels, kernelSize, stride, numFilters, padding,
activationType, input, kernels.data(), d_input
);
EXPECT_EQ(inputSize, conv2d.getOutputSize());
@@ -203,13 +205,14 @@ TEST_F(Conv2dTest, PaddedTest) {
}
TEST_F(Conv2dTest, StridedPaddedConvolution) {
int inputSize = 5;
int inputChannels = 2;
int kernelSize = 3;
int stride = 2;
int numFilters = 2;
CUDANet::Layers::Padding padding = CUDANet::Layers::Padding::SAME;
CUDANet::Layers::Activation activation = CUDANet::Layers::Activation::RELU;
int inputSize = 5;
int inputChannels = 2;
int kernelSize = 3;
int stride = 2;
int numFilters = 2;
CUDANet::Layers::Padding padding = CUDANet::Layers::Padding::SAME;
CUDANet::Layers::ActivationType activationType =
CUDANet::Layers::ActivationType::RELU;
// clang-format off
std::vector<float> input = {
@@ -250,8 +253,8 @@ TEST_F(Conv2dTest, StridedPaddedConvolution) {
float* d_output;
CUDANet::Layers::Conv2d conv2d = commonTestSetup(
inputSize, inputChannels, kernelSize, stride, padding, numFilters,
activation, input, kernels.data(), d_input
inputSize, inputChannels, kernelSize, stride, numFilters, padding,
activationType, input, kernels.data(), d_input
);
EXPECT_EQ(inputSize, conv2d.getOutputSize());

16
test/layers/test_dense.cu
View File

@@ -3,7 +3,7 @@
#include <iostream>
#include "activations.cuh"
#include "activation.cuh"
#include "dense.cuh"
class DenseLayerTest : public ::testing::Test {
@@ -15,10 +15,10 @@ class DenseLayerTest : public ::testing::Test {
float* weights,
float* biases,
float*& d_input,
CUDANet::Layers::Activation activation
CUDANet::Layers::ActivationType activationType
) {
// Create Dense layer
CUDANet::Layers::Dense denseLayer(inputSize, outputSize, activation);
CUDANet::Layers::Dense denseLayer(inputSize, outputSize, activationType);
// Set weights and biases
denseLayer.setWeights(weights);
@@ -53,7 +53,7 @@ TEST_F(DenseLayerTest, Init) {
int outputSize = j;
CUDANet::Layers::Dense denseLayer(
inputSize, outputSize, CUDANet::Layers::Activation::SIGMOID
inputSize, outputSize, CUDANet::Layers::ActivationType::SIGMOID
);
}
}
@@ -74,7 +74,7 @@ TEST_F(DenseLayerTest, setWeights) {
// clang-format on
CUDANet::Layers::Dense denseLayer(
inputSize, outputSize, CUDANet::Layers::Activation::SIGMOID
inputSize, outputSize, CUDANet::Layers::ActivationType::SIGMOID
);
denseLayer.setWeights(weights.data());
@@ -101,7 +101,7 @@ TEST_F(DenseLayerTest, ForwardUnitWeightMatrixLinear) {
CUDANet::Layers::Dense denseLayer = commonTestSetup(
inputSize, outputSize, input, weights.data(), biases.data(), d_input,
CUDANet::Layers::Activation::NONE
CUDANet::Layers::ActivationType::NONE
);
d_output = denseLayer.forward(d_input);
@@ -142,7 +142,7 @@ TEST_F(DenseLayerTest, ForwardRandomWeightMatrixRelu) {
CUDANet::Layers::Dense denseLayer = commonTestSetup(
inputSize, outputSize, input, weights.data(), biases.data(), d_input,
CUDANet::Layers::Activation::RELU
CUDANet::Layers::ActivationType::RELU
);
d_output = denseLayer.forward(d_input);
@@ -187,7 +187,7 @@ TEST_F(DenseLayerTest, ForwardRandomWeightMatrixSigmoid) {
CUDANet::Layers::Dense denseLayer = commonTestSetup(
inputSize, outputSize, input, weights.data(), biases.data(), d_input,
CUDANet::Layers::Activation::SIGMOID
CUDANet::Layers::ActivationType::SIGMOID
);
d_output = denseLayer.forward(d_input);