Relintai/MLPP
1
0
Fork 0
mirror of https://github.com/Relintai/MLPP.git synced 2024-11-12 10:15:01 +01:00
Code Issues Packages Projects Releases Wiki Activity

explicitly added sgd function in ann

Browse Source
This commit is contained in:
novak_99 2022-02-12 11:16:22 -08:00
parent 4cc61e4c1e
commit 2a21d25999
5 changed files with 64 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

35
MLPP/ANN/ANN.cpp
View File

@ -13,6 +13,7 @@
#include <iostream>
#include <cmath>
#include <random>
namespace MLPP {
ANN::ANN(std::vector<std::vector<double>> inputSet, std::vector<double> outputSet)
@ -87,6 +88,40 @@ namespace MLPP {
}
}
void ANN::SGD(double learning_rate, int max_epoch, bool UI){
class Cost cost;
LinAlg alg;
double cost_prev = 0;
int epoch = 1;
double initial_learning_rate = learning_rate;
while(true){
learning_rate = applyLearningRateScheduler(initial_learning_rate, decayConstant, epoch, dropRate);
std::random_device rd;
std::default_random_engine generator(rd());
std::uniform_int_distribution<int> distribution(0, int(n - 1));
int outputIndex = distribution(generator);
std::vector<double> y_hat = modelSetTest({inputSet[outputIndex]});
cost_prev = Cost({y_hat}, {outputSet[outputIndex]});
auto [cumulativeHiddenLayerWGrad, outputWGrad] = computeGradients(y_hat, {outputSet[outputIndex]});
cumulativeHiddenLayerWGrad = alg.scalarMultiply(learning_rate/n, cumulativeHiddenLayerWGrad);
outputWGrad = alg.scalarMultiply(learning_rate/n, outputWGrad);
updateParameters(cumulativeHiddenLayerWGrad, outputWGrad, learning_rate); // subject to change. may want bias to have this matrix too.
y_hat = modelSetTest({inputSet[outputIndex]});
if(UI) { ANN::UI(epoch, cost_prev, y_hat, {outputSet[outputIndex]}); }
epoch++;
if(epoch > max_epoch) { break; }
}
forwardPass();
}
void ANN::MBGD(double learning_rate, int max_epoch, int mini_batch_size, bool UI){
class Cost cost;
LinAlg alg;

1
MLPP/ANN/ANN.hpp
View File

@ -23,6 +23,7 @@ class ANN{
std::vector<double> modelSetTest(std::vector<std::vector<double>> X);
double modelTest(std::vector<double> x);
void gradientDescent(double learning_rate, int max_epoch, bool UI = 1);
void SGD(double learning_rate, int max_epoch, bool UI = 1);
void MBGD(double learning_rate, int max_epoch, int mini_batch_size, bool UI = 1);
void Momentum(double learning_rate, int max_epoch, int mini_batch_size, double gamma, bool NAG, bool UI = 1);
void Adagrad(double learning_rate, int max_epoch, int mini_batch_size, double e, bool UI = 1);

1
MLPP/Transforms/Transforms.cpp
View File

@ -13,6 +13,7 @@
namespace MLPP{
// DCT ii.
// https://www.mathworks.com/help/images/discrete-cosine-transform.html
std::vector<std::vector<double>> Transforms::discreteCosineTransform(std::vector<std::vector<double>> A){
LinAlg alg;
A = alg.scalarAdd(-128, A); // Center around 0.

BIN
a.out
View File

Binary file not shown.

50
main.cpp
View File

@ -363,20 +363,22 @@ int main() {
// Possible Weight Init Methods: Default, Uniform, HeNormal, HeUniform, XavierNormal, XavierUniform
// Possible Activations: Linear, Sigmoid, Swish, Softplus, Softsign, CLogLog, Ar{Sinh, Cosh, Tanh, Csch, Sech, Coth}, GaussianCDF, GELU, UnitStep
// Possible Loss Functions: MSE, RMSE, MBE, LogLoss, CrossEntropy, HingeLoss
// std::vector<std::vector<double>> inputSet = {{0,0,1,1}, {0,1,0,1}};
// std::vector<double> outputSet = {0,1,1,0};
// ANN ann(alg.transpose(inputSet), outputSet);
// ann.addLayer(2, "Sigmoid");
// ann.addLayer(2, "Sigmoid");
// ann.addOutputLayer("Sigmoid", "LogLoss");
//ann.AMSGrad(0.1, 10000, 1, 0.9, 0.999, 0.000001, 1);
//ann.Adadelta(1, 1000, 2, 0.9, 0.000001, 1);
//ann.Momentum(0.1, 8000, 2, 0.9, true, 1);
std::vector<std::vector<double>> inputSet = {{0,0,1,1}, {0,1,0,1}};
std::vector<double> outputSet = {0,1,1,0};
ANN ann(alg.transpose(inputSet), outputSet);
ann.addLayer(2, "Sigmoid");
ann.addLayer(2, "Sigmoid");
ann.addOutputLayer("Sigmoid", "LogLoss");
// ann.AMSGrad(0.1, 10000, 1, 0.9, 0.999, 0.000001, 1);
// ann.Adadelta(1, 1000, 2, 0.9, 0.000001, 1);
// ann.Momentum(0.1, 8000, 2, 0.9, true, 1);
//ann.setLearningRateScheduler("Step", 0.5, 1000);
// ann.gradientDescent(1, 5, 1);
//alg.printVector(ann.modelSetTest(alg.transpose(inputSet)));
//std::cout << "ACCURACY: " << 100 * ann.score() << "%" << std::endl;
ann.SGD(1, 30000);
alg.printVector(ann.modelSetTest(alg.transpose(inputSet)));
std::cout << "ACCURACY: " << 100 * ann.score() << "%" << std::endl;
//std::vector<std::vector<double>> outputSet = {{1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20},
// {2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38,40}};
@ -477,20 +479,20 @@ int main() {
// alg.printTensor(data.rgb2xyz(tensorSet));
std::vector<std::vector<double>> input = {
{62,55,55,54,49,48,47,55},
{62,57,54,52,48,47,48,53},
{61,60,52,49,48,47,49,54},
{63,61,60,60,63,65,68,65},
{67,67,70,74,79,85,91,92},
{82,95,101,106,114,115,112,117},
{96,111,115,119,128,128,130,127},
{109,121,127,133,139,141,140,133},
};
// std::vector<std::vector<double>> input = {
// {62,55,55,54,49,48,47,55},
// {62,57,54,52,48,47,48,53},
// {61,60,52,49,48,47,49,54},
// {63,61,60,60,63,65,68,65},
// {67,67,70,74,79,85,91,92},
// {82,95,101,106,114,115,112,117},
// {96,111,115,119,128,128,130,127},
// {109,121,127,133,139,141,140,133},
// };
Transforms trans;
// Transforms trans;
alg.printMatrix(trans.discreteCosineTransform(input));
// alg.printMatrix(trans.discreteCosineTransform(input));
// alg.printMatrix(conv.convolve(input, conv.getPrewittVertical(), 1)); // Can use padding
// alg.printMatrix(conv.pool(input, 4, 4, "Max")); // Can use Max, Min, or Average pooling.