shin/NeuralNetworkLib
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Rprop implementation

Browse Source
This commit is contained in:
Shin
2016-10-30 23:00:50 +01:00
parent 554ef1b46b
commit 8749b3eb03
5 changed files with 415 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
tests/CMakeLists.txt
View File

@@ -28,6 +28,9 @@ target_link_libraries(recurrent NeuralNetwork gtest gtest_main)
add_executable(quickpropagation quickpropagation.cpp)
target_link_libraries(quickpropagation NeuralNetwork gtest gtest_main)
add_executable(rprop rprop.cpp)
target_link_libraries(rprop NeuralNetwork gtest gtest_main)
# PERF
add_executable(backpropagation_function_cmp backpropagation_function_cmp.cpp)

165
tests/rprop.cpp Normal file
View File

@@ -0,0 +1,165 @@
#include <NeuralNetwork/FeedForward/Network.h>
#include <NeuralNetwork/Learning/RProp.h>
#include <NeuralNetwork/ActivationFunction/HyperbolicTangent.h>
#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Weffc++"
#include <gtest/gtest.h>
#pragma GCC diagnostic pop
TEST(RProp,XOR) {
NeuralNetwork::FeedForward::Network n(2);
NeuralNetwork::ActivationFunction::Sigmoid a(-1);
n.appendLayer(3,a);
n.appendLayer(1,a);
n.randomizeWeights();
NeuralNetwork::Learning::RProp prop(n);
prop.setBatchSize(4);
for(int i=0;i<100;i++) {
prop.teach({1,0},{1});
prop.teach({1,1},{0});
prop.teach({0,0},{0});
prop.teach({0,1},{1});
}
{
std::vector<float> ret =n.computeOutput({1,1});
ASSERT_LT(ret[0], 0.1);
}
{
std::vector<float> ret =n.computeOutput({0,1});
ASSERT_GT(ret[0], 0.9);
}
{
std::vector<float> ret =n.computeOutput({1,0});
ASSERT_GT(ret[0], 0.9);
}
{
std::vector<float> ret =n.computeOutput({0,0});
ASSERT_LT(ret[0], 0.1);
}
}
TEST(RProp,XORHyperbolicTangent) {
srand(time(NULL));
NeuralNetwork::FeedForward::Network n(2);
NeuralNetwork::ActivationFunction::HyperbolicTangent a(-1);
n.appendLayer(2,a);
n.appendLayer(1,a);
n.randomizeWeights();
NeuralNetwork::Learning::RProp prop(n);
prop.setBatchSize(4);
for(int i=0;i<15000;i++) {
prop.teach({1,0},{1});
prop.teach({1,1},{0});
prop.teach({0,0},{0});
prop.teach({0,1},{1});
}
{
std::vector<float> ret =n.computeOutput({1,1});
ASSERT_LT(ret[0], 0.1);
}
{
std::vector<float> ret =n.computeOutput({0,1});
ASSERT_GT(ret[0], 0.9);
}
{
std::vector<float> ret =n.computeOutput({1,0});
ASSERT_GT(ret[0], 0.9);
}
{
std::vector<float> ret =n.computeOutput({0,0});
ASSERT_LT(ret[0], 0.1);
}
}
TEST(RProp,AND) {
NeuralNetwork::FeedForward::Network n(2);
NeuralNetwork::ActivationFunction::Sigmoid a(-1);
n.appendLayer(1,a);
n.randomizeWeights();
NeuralNetwork::Learning::RProp prop(n);
prop.setBatchSize(4);
for(int i=0;i<100000;i++) {
prop.teach({1,1},{1});
prop.teach({0,0},{0});
prop.teach({0,1},{0});
prop.teach({1,0},{0});
}
{
std::vector<float> ret =n.computeOutput({1,1});
ASSERT_GT(ret[0], 0.9);
}
{
std::vector<float> ret =n.computeOutput({0,1});
ASSERT_LT(ret[0], 0.1);
}
{
std::vector<float> ret =n.computeOutput({1,0});
ASSERT_LT(ret[0], 0.1);
}
{
std::vector<float> ret =n.computeOutput({0,0});
ASSERT_LT(ret[0], 0.1);
}
}
TEST(RProp,NOTAND) {
NeuralNetwork::FeedForward::Network n(2);
NeuralNetwork::ActivationFunction::Sigmoid a(-1);
n.appendLayer(2,a);
n.appendLayer(1,a);
n.randomizeWeights();
NeuralNetwork::Learning::RProp prop(n);
prop.setBatchSize(4);
for(int i=0;i<100000;i++) {
prop.teach({1,1},{0});
prop.teach({0,0},{1});
prop.teach({0,1},{1});
prop.teach({1,0},{1});
}
{
std::vector<float> ret =n.computeOutput({1,1});
ASSERT_LT(ret[0], 0.1);
}
{
std::vector<float> ret =n.computeOutput({0,1});
ASSERT_GT(ret[0], 0.9);
}
{
std::vector<float> ret =n.computeOutput({1,0});
ASSERT_GT(ret[0], 0.9);
}
{
std::vector<float> ret =n.computeOutput({0,0});
ASSERT_GT(ret[0], 0.9);
}
}