157 lines
5.7 KiB
C++
157 lines
5.7 KiB
C++
|
|
#include "gtest/gtest.h"
|
|
|
|
#include "kompute/Kompute.hpp"
|
|
|
|
#include "kompute_test/shaders/shadertest_logistic_regression.hpp"
|
|
|
|
TEST(TestLogisticRegression, TestMainLogisticRegression)
|
|
{
|
|
|
|
uint32_t ITERATIONS = 100;
|
|
float learningRate = 0.1;
|
|
|
|
{
|
|
kp::Manager mgr;
|
|
|
|
std::shared_ptr<kp::Tensor> xI = mgr.tensor({ 0, 1, 1, 1, 1 });
|
|
std::shared_ptr<kp::Tensor> xJ = mgr.tensor({ 0, 0, 0, 1, 1 });
|
|
|
|
std::shared_ptr<kp::Tensor> y = mgr.tensor({ 0, 0, 0, 1, 1 });
|
|
|
|
std::shared_ptr<kp::Tensor> wIn = mgr.tensor({ 0.001, 0.001 });
|
|
std::shared_ptr<kp::Tensor> wOutI = mgr.tensor({ 0, 0, 0, 0, 0 });
|
|
std::shared_ptr<kp::Tensor> wOutJ = mgr.tensor({ 0, 0, 0, 0, 0 });
|
|
|
|
std::shared_ptr<kp::Tensor> bIn = mgr.tensor({ 0 });
|
|
std::shared_ptr<kp::Tensor> bOut = mgr.tensor({ 0, 0, 0, 0, 0 });
|
|
|
|
std::shared_ptr<kp::Tensor> lOut = mgr.tensor({ 0, 0, 0, 0, 0 });
|
|
|
|
std::vector<std::shared_ptr<kp::Tensor>> params = { xI, xJ, y,
|
|
wIn, wOutI, wOutJ,
|
|
bIn, bOut, lOut };
|
|
|
|
mgr.sequence()->eval<kp::OpTensorSyncDevice>(params);
|
|
|
|
std::vector<uint32_t> spirv = std::vector<uint32_t>(
|
|
(uint32_t*)kp::shader_data::
|
|
test_shaders_glsl_test_logistic_regression_comp_spv,
|
|
(uint32_t*)(kp::shader_data::
|
|
test_shaders_glsl_test_logistic_regression_comp_spv +
|
|
kp::shader_data::
|
|
test_shaders_glsl_test_logistic_regression_comp_spv_len));
|
|
|
|
std::shared_ptr<kp::Algorithm> algorithm = mgr.algorithm(
|
|
params, spirv, kp::Workgroup({ 5 }), kp::Constants({ 5.0 }));
|
|
|
|
std::shared_ptr<kp::Sequence> sq =
|
|
mgr.sequence()
|
|
->record<kp::OpTensorSyncDevice>({ wIn, bIn })
|
|
->record<kp::OpAlgoDispatch>(algorithm)
|
|
->record<kp::OpTensorSyncLocal>({ wOutI, wOutJ, bOut, lOut });
|
|
|
|
// Iterate across all expected iterations
|
|
for (size_t i = 0; i < ITERATIONS; i++) {
|
|
|
|
sq->eval();
|
|
|
|
for (size_t j = 0; j < bOut->size(); j++) {
|
|
wIn->data()[0] -= learningRate * wOutI->data()[j];
|
|
wIn->data()[1] -= learningRate * wOutJ->data()[j];
|
|
bIn->data()[0] -= learningRate * bOut->data()[j];
|
|
}
|
|
}
|
|
|
|
// Based on the inputs the outputs should be at least:
|
|
// * wi < 0.01
|
|
// * wj > 1.0
|
|
// * b < 0
|
|
// TODO: Add EXPECT_DOUBLE_EQ instead
|
|
EXPECT_LT(wIn->data()[0], 0.01);
|
|
EXPECT_GT(wIn->data()[1], 1.0);
|
|
EXPECT_LT(bIn->data()[0], 0.0);
|
|
|
|
KP_LOG_WARN("Result wIn i: {}, wIn j: {}, bIn: {}",
|
|
wIn->data()[0],
|
|
wIn->data()[1],
|
|
bIn->data()[0]);
|
|
}
|
|
}
|
|
|
|
TEST(TestLogisticRegression, TestMainLogisticRegressionManualCopy)
|
|
{
|
|
|
|
uint32_t ITERATIONS = 100;
|
|
float learningRate = 0.1;
|
|
|
|
{
|
|
kp::Manager mgr;
|
|
|
|
std::shared_ptr<kp::Tensor> xI = mgr.tensor({ 0, 1, 1, 1, 1 });
|
|
std::shared_ptr<kp::Tensor> xJ = mgr.tensor({ 0, 0, 0, 1, 1 });
|
|
|
|
std::shared_ptr<kp::Tensor> y = mgr.tensor({ 0, 0, 0, 1, 1 });
|
|
|
|
std::shared_ptr<kp::Tensor> wIn =
|
|
mgr.tensor({ 0.001, 0.001 }, kp::Tensor::TensorTypes::eHost);
|
|
std::shared_ptr<kp::Tensor> wOutI = mgr.tensor({ 0, 0, 0, 0, 0 });
|
|
std::shared_ptr<kp::Tensor> wOutJ = mgr.tensor({ 0, 0, 0, 0, 0 });
|
|
|
|
std::shared_ptr<kp::Tensor> bIn =
|
|
mgr.tensor({ 0 }, kp::Tensor::TensorTypes::eHost);
|
|
std::shared_ptr<kp::Tensor> bOut = mgr.tensor({ 0, 0, 0, 0, 0 });
|
|
|
|
std::shared_ptr<kp::Tensor> lOut = mgr.tensor({ 0, 0, 0, 0, 0 });
|
|
|
|
std::vector<std::shared_ptr<kp::Tensor>> params = { xI, xJ, y,
|
|
wIn, wOutI, wOutJ,
|
|
bIn, bOut, lOut };
|
|
|
|
mgr.sequence()->record<kp::OpTensorSyncDevice>(params)->eval();
|
|
|
|
std::vector<uint32_t> spirv = std::vector<uint32_t>(
|
|
(uint32_t*)kp::shader_data::shaders_glsl_logisticregression_comp_spv,
|
|
(uint32_t*)(kp::shader_data::
|
|
shaders_glsl_logisticregression_comp_spv +
|
|
kp::shader_data::
|
|
shaders_glsl_logisticregression_comp_spv_len));
|
|
|
|
std::shared_ptr<kp::Algorithm> algorithm =
|
|
mgr.algorithm(params, spirv, kp::Workgroup(), kp::Constants({ 5.0 }));
|
|
|
|
std::shared_ptr<kp::Sequence> sq =
|
|
mgr.sequence()
|
|
->record<kp::OpTensorSyncDevice>({ wIn, bIn })
|
|
->record<kp::OpAlgoDispatch>(algorithm)
|
|
->record<kp::OpTensorSyncLocal>({ wOutI, wOutJ, bOut, lOut });
|
|
|
|
// Iterate across all expected iterations
|
|
for (size_t i = 0; i < ITERATIONS; i++) {
|
|
|
|
sq->eval();
|
|
|
|
for (size_t j = 0; j < bOut->size(); j++) {
|
|
wIn->data()[0] -= learningRate * wOutI->data()[j];
|
|
wIn->data()[1] -= learningRate * wOutJ->data()[j];
|
|
bIn->data()[0] -= learningRate * bOut->data()[j];
|
|
}
|
|
wIn->mapDataIntoHostMemory();
|
|
bIn->mapDataIntoHostMemory();
|
|
}
|
|
|
|
// Based on the inputs the outputs should be at least:
|
|
// * wi < 0.01
|
|
// * wj > 1.0
|
|
// * b < 0
|
|
// TODO: Add EXPECT_DOUBLE_EQ instead
|
|
EXPECT_LT(wIn->data()[0], 0.01);
|
|
EXPECT_GT(wIn->data()[1], 1.0);
|
|
EXPECT_LT(bIn->data()[0], 0.0);
|
|
|
|
KP_LOG_WARN("Result wIn i: {}, wIn j: {}, bIn: {}",
|
|
wIn->data()[0],
|
|
wIn->data()[1],
|
|
bIn->data()[0]);
|
|
}
|
|
}
|