don't know

src/Cuda/Makefile

@@ -0,0 +1,8 @@
+all: test
+
+VectorOperations.o: VectorOperations.cu VectorOperations.h
+	/usr/local/cuda-6.5/bin/nvcc -c VectorOperations.cu -o VectorOperations.o
+
+test: VectorOperations.o VectorOperations.h test.cpp
+	g++  -c -std=c++14 -O0 ./test.cpp -o test.o
+	/usr/local/cuda-6.5/bin/nvcc ./test.o VectorOperations.o -o test

src/Cuda/VectorOperations.cpp

@@ -0,0 +1,12 @@
+#include "./VectorOperations.h"
+
+// CUDA kernel. Each thread takes care of one element of c
+__global__ void vecAdd(double *a, double *b, double *c, int n)
+{
+    // Get our global thread ID
+    int id = blockIdx.x*blockDim.x+threadIdx.x;
+ 
+    // Make sure we do not go out of bounds
+    if (id < n)
+        c[id] = a[id] + b[id];
+}

src/Cuda/VectorOperations.cu

@@ -0,0 +1,47 @@
+#include "./VectorOperations.h"
+
+// CUDA kernel. Each thread takes care of one element of c
+__global__ void __ADD(float *a, float *b, float *ret, int n)
+{
+    // Get our global thread ID
+    int id = blockIdx.x*blockDim.x+threadIdx.x;
+ 
+    if (id >= n)
+		return;
+
+	// Make sure we do not go out of bounds
+	float tmp=0.0;
+	for(int i=0;i<n;i++)
+	{
+		tmp+=a[n] * b[n];
+	}
+    ret[id] = tmp;
+}
+
+Shin::Cuda::VectorOperations::VectorOperations(int size)
+{
+	cudaMalloc(&clientA, sizeof(float)*size);
+	cudaMalloc(&clientB, sizeof(float)*size);
+	cudaMalloc(&clientC, sizeof(float)*size);
+}
+
+Shin::Cuda::VectorOperations::~VectorOperations()
+{
+	cudaFree(clientA);
+	cudaFree(clientB);
+	cudaFree(clientC);
+}
+
+void Shin::Cuda::VectorOperations::add(float *a, float *b, float *ret, int n)
+{
+	//cudaMemcpyAsync(clientA, a, n*sizeof(float), cudaMemcpyHostToDevice, );
+	//cudaMemcpyAsync(clientB, a, n*sizeof(float), cudaMemcpyHostToDevice, );
+	cudaMemcpy( clientA, a, n*sizeof(float), cudaMemcpyHostToDevice);
+	cudaMemcpy( clientB, b, n*sizeof(float), cudaMemcpyHostToDevice);
+	cudaMemcpy( clientC, ret, n*sizeof(float), cudaMemcpyHostToDevice);
+	int blockSize, gridSize;
+	blockSize = 1024;
+	gridSize = (int)ceil((float)n/blockSize);
+	__ADD<<< 1, n>>>(clientA, clientB, clientC, n);
+	cudaMemcpy( ret, clientC, n*sizeof(float), cudaMemcpyDeviceToHost );
+}

src/Cuda/VectorOperations.h

@@ -0,0 +1,20 @@
+#ifndef VECT_OP_
+#define VECT_OP_
+namespace Shin
+{
+namespace Cuda
+{
+	class VectorOperations
+	{
+		public:
+			VectorOperations(int maxSize);
+			~VectorOperations();
+			void add(float *a, float *b, float *ret, int n);
+		protected:
+			float *clientA;
+			float *clientB;
+			float *clientC;
+	};
+}
+}
+#endif

src/Cuda/answer.cu

@@ -0,0 +1,66 @@
+#include <stdio.h>
+
+__global__ void vector_add(int *a, int *b, int *c)
+{
+    /* insert code to calculate the index properly using blockIdx.x, blockDim.x, threadIdx.x */
+	int index = blockIdx.x * blockDim.x + threadIdx.x;
+	c[index] = a[index] + b[index];
+}
+
+/* experiment with N */
+/* how large can it be? */
+#define N (2048*2048)
+#define THREADS_PER_BLOCK 512
+
+int main()
+{
+    int *a, *b, *c;
+	int *d_a, *d_b, *d_c;
+	int size = N * sizeof( int );
+
+	/* allocate space for device copies of a, b, c */
+
+	cudaMalloc( (void **) &d_a, size );
+	cudaMalloc( (void **) &d_b, size );
+	cudaMalloc( (void **) &d_c, size );
+
+	/* allocate space for host copies of a, b, c and setup input values */
+
+	a = (int *)malloc( size );
+	b = (int *)malloc( size );
+	c = (int *)malloc( size );
+
+	for( int i = 0; i < N; i++ )
+	{
+		a[i] = b[i] = i;
+		c[i] = 0;
+	}
+
+	/* copy inputs to device */
+	/* fix the parameters needed to copy data to the device */
+	cudaMemcpy( d_a, a, size, cudaMemcpyHostToDevice );
+	cudaMemcpy( d_b, b, size, cudaMemcpyHostToDevice );
+
+	/* launch the kernel on the GPU */
+	/* insert the launch parameters to launch the kernel properly using blocks and threads */ 
+	vector_add<<< (N + (THREADS_PER_BLOCK-1)) / THREADS_PER_BLOCK, THREADS_PER_BLOCK >>>( d_a, d_b, d_c );
+
+	/* copy result back to host */
+	/* fix the parameters needed to copy data back to the host */
+	cudaMemcpy( c, d_c, size, cudaMemcpyDeviceToHost );
+
+
+	printf( "c[0] = %d\n",0,c[0] );
+	printf( "c[%d] = %d\n",N-1, c[N-1] );
+
+	/* clean up */
+
+	free(a);
+	free(b);
+	free(c);
+	cudaFree( d_a );
+	cudaFree( d_b );
+	cudaFree( d_c );
+	
+	return 0;
+} /* end main */

src/Cuda/answer.cu~

@@ -0,0 +1,66 @@
+#include <stdio.h>
+
+__global__ void vector_add(int *a, int *b, int *c)
+{
+    /* insert code to calculate the index properly using blockIdx.x, blockDim.x, threadIdx.x */
+	int index = blockIdx.x * blockDim.x + threadIdx.x;
+	c[index] = a[index] + b[index];
+}
+
+/* experiment with N */
+/* how large can it be? */
+#define N (2048*2048)
+#define THREADS_PER_BLOCK 512
+
+int main()
+{
+    int *a, *b, *c;
+	int *d_a, *d_b, *d_c;
+	int size = N * sizeof( int );
+
+	/* allocate space for device copies of a, b, c */
+
+	cudaMalloc( (void **) &d_a, size );
+	cudaMalloc( (void **) &d_b, size );
+	cudaMalloc( (void **) &d_c, size );
+
+	/* allocate space for host copies of a, b, c and setup input values */
+
+	a = (int *)malloc( size );
+	b = (int *)malloc( size );
+	c = (int *)malloc( size );
+
+	for( int i = 0; i < N; i++ )
+	{
+		a[i] = b[i] = i;
+		c[i] = 0;
+	}
+
+	/* copy inputs to device */
+	/* fix the parameters needed to copy data to the device */
+	cudaMemcpy( d_a, a, size, cudaMemcpyHostToDevice );
+	cudaMemcpy( d_b, b, size, cudaMemcpyHostToDevice );
+
+	/* launch the kernel on the GPU */
+	/* insert the launch parameters to launch the kernel properly using blocks and threads */ 
+	add<<< (N + (THREADS_PER_BLOCK-1)) / THREADS_PER_BLOCK, THREADS_PER_BLOCK >>>( d_a, d_b, d_c );
+
+	/* copy result back to host */
+	/* fix the parameters needed to copy data back to the host */
+	cudaMemcpy( c, d_c, size, cudaMemcpyDeviceToHost );
+
+
+	printf( "c[0] = %d\n",0,c[0] );
+	printf( "c[%d] = %d\n",N-1, c[N-1] );
+
+	/* clean up */
+
+	free(a);
+	free(b);
+	free(c);
+	cudaFree( d_a );
+	cudaFree( d_b );
+	cudaFree( d_c );
+	
+	return 0;
+} /* end main */

src/Cuda/test.cpp

@@ -0,0 +1,44 @@
+#include "./VectorOperations.h"
+#include <chrono>
+#include <iostream>
+void _hack(float *a)
+{
+	a[0]=a[0];
+}
+int main()
+{
+	int size=50000000;
+	float *a= new float[size];
+	float *b= new float[size];
+	float *c= new float[size];
+	double sum = 0;
+	Shin::Cuda::VectorOperations v(size);
+	for(int i=0;i<size;i++)
+	{
+		a[i]=0.001;
+		b[i]=2;
+		c[i]=0;
+	}
+
+	_hack(a);
+    auto t1 = std::chrono::high_resolution_clock::now();
+
+	for(int i=0;i<size;i++)
+	{
+		sum+=a[i]*b[i];
+	}
+
+    auto t2 = std::chrono::high_resolution_clock::now();
+
+	std::cout << "Time 1: " << std::chrono::duration_cast<std::chrono::milliseconds>(t2-t1).count() << "result: " << sum <<std::endl;
+
+	for(int i=0;i<size;i++)
+		c[i]=0;
+	t1 = std::chrono::high_resolution_clock::now();
+	v.add(a,b,c,size);
+	sum=0;
+	for(int i=0;i<size;i++)
+		sum+=c[i];
+    t2 = std::chrono::high_resolution_clock::now();
+	std::cout << "Time 2: " << std::chrono::duration_cast<std::chrono::milliseconds>(t2-t1).count() << "result: " << sum << std::endl;
+}

src/Makefile

@@ -0,0 +1,8 @@
+OBJFILES= sse_mathfun.o
+
+include ../Makefile.const
+
+all: $(OBJFILES)
+
+clean:
+	@rm -f ./*.o ./*.so ./*.a ./*.nm ./*/*.o

src/NeuronNetwork/FeedForwardQuick.cpp

@@ -122,13 +122,14 @@ void FeedForwardNetworkQuick::solvePart(float *newSolution, register size_t begi
 
 Solution FeedForwardNetworkQuick::solve(const Problem& p)
 {
-	std::vector<float> solution(p);
 	register float* sol=sums[1];
 
-	for(size_t i=0;i<solution.size();i++)
+	sums[0][0]=1;
+	sol[0]=1;
+	for(size_t i=0;i<p.size();i++)
 	{
-		sums[0][i+1]=solution[i];
+		sums[0][i+1]=p[i];
-		sol[i+1]=solution[i];
+		sol[i+1]=p[i];
 	}
 
 	register size_t prevSize=layerSizes[0];
@@ -160,7 +161,7 @@ Solution FeedForwardNetworkQuick::solve(const Problem& p)
 		prevSize=layerSizes[i];
 		sol=newSolution;
 	}
-	std::vector<float> ret;
+	Solution ret;
 	for(size_t i=1;i<prevSize;i++)
 	{
 		ret.push_back(sol[i]);

src/NeuronNetwork/IO.cpp

@@ -20,7 +20,7 @@ Shin::NeuronNetwork::IO::~IO()
 
 }
 
-Shin::NeuronNetwork::IO Shin::NeuronNetwork::IO::operator+(IO &r)
+Shin::NeuronNetwork::IO Shin::NeuronNetwork::IO::operator+(const IO &r)
 {
 	Shin::NeuronNetwork::IO tmp;
 	for(float a:this->data)
@@ -39,6 +39,11 @@ Shin::NeuronNetwork::IO::operator std::vector<float>&()
 	return data;
 }
 
+Shin::NeuronNetwork::IO::operator std::vector<float>()
+{
+	return data;
+}
+
 
 float Shin::NeuronNetwork::IO::operator[](size_t pos) const
 {

src/NeuronNetwork/IO.h

@@ -15,8 +15,9 @@ class IO
 		IO(std::vector<float> &d);
 		IO(const IO &old);
 		virtual ~IO();
-		IO operator+(IO &r);
+		IO operator+(const IO &r);
 		virtual operator std::vector<float>&() final; // TOO CONST
+		virtual operator std::vector<float>() final; // TOO CONST
 		float operator[] (size_t pos) const;
 		size_t size() const;
 	protected:

src/NeuronNetwork/Learning/BackPropagation.cpp

@@ -99,7 +99,7 @@ float Shin::NeuronNetwork::Learning::BackPropagation::teach(const Shin::NeuronNe
 	Shin::NeuronNetwork::Solution a=network.solve(p);
 	double error=calculateError(solution,a);
 
-	std::vector<float> s;
+	Solution s;
 	if(entropy)
 	{
 		for(size_t i=0;i<solution.size();i++)

src/NeuronNetwork/Learning/QLearning

@@ -0,0 +1 @@
+./QLearning.h

src/NeuronNetwork/Learning/QLearning.cpp

@@ -0,0 +1,38 @@
+#include "./QLearning"
+
+Shin::NeuronNetwork::Learning::QLearning::QLearning(size_t input, size_t size, size_t choices):fun()
+{
+	fun.initialiseNetwork(input,size,choices);
+}
+
+Shin::NeuronNetwork::Learning::QLearning::~QLearning()
+{
+
+}
+
+void Shin::NeuronNetwork::Learning::QLearning::learnDelayed(std::vector< std::pair< Shin::NeuronNetwork::Solution, Shin::NeuronNetwork::Problem > >& p, float quality)
+{
+	fun.learnDelayed(p,quality);
+}
+
+void Shin::NeuronNetwork::Learning::QLearning::learnDelayed(std::vector< std::pair< Shin::NeuronNetwork::Problem,int > >& p, float quality)
+{
+	fun.learnDelayed(p,quality);
+}
+
+void Shin::NeuronNetwork::Learning::QLearning::learn(Shin::NeuronNetwork::Solution& s, Shin::NeuronNetwork::Problem& p, float quality)
+{
+	fun.learn(s,p,quality);
+}
+
+
+void Shin::NeuronNetwork::Learning::QLearning::learn(Shin::NeuronNetwork::Problem& s, int action, float quality)
+{
+	fun.learn(s,action,quality);
+}
+
+
+int Shin::NeuronNetwork::Learning::QLearning::getChoice(Shin::NeuronNetwork::Problem& p)
+{
+	return fun.getChoice(p);
+}

src/NeuronNetwork/Learning/QLearning.h

@@ -0,0 +1,70 @@
+#ifndef _QLEARNING_H_
+#define _QLEARNING_H_
+
+#include <math.h>
+#include <cstddef>
+
+#include "../Problem.h"
+#include "../FeedForwardQuick.h"
+#include "BackPropagation"
+#include  "Unsupervised"
+#include "RL/QFunction.h"
+#include "OpticalBackPropagation.h"
+#include "functional"
+
+/*
+ * http://www2.econ.iastate.edu/tesfatsi/RLUsersGuide.ICAC2005.pdf
+ * http://www.autonlab.org/tutorials/rl06.pdf
+ * http://www.nbu.bg/cogs/events/2000/Readings/Petrov/rltutorial.pdf
+ * 
+ * http://www.applied-mathematics.net/qlearning/qlearning.html
+ * http://nn.cs.utexas.edu/downloads/papers/stanley.gecco02_1.pdf
+ * 
+ * http://stackoverflow.com/questions/740389/good-implementations-of-reinforced-learning
+ * 
+ * http://stackoverflow.com/questions/10722064/training-a-neural-network-with-reinforcement-learning
+ * 
+ * http://remi.coulom.free.fr/Thesis/
+ * http://remi.coulom.free.fr/Publications/Thesis.pdf
+ * 
+ * http://link.springer.com/article/10.1007/BF00992696
+ * 
+ * http://scholar.google.cz/scholar?start=10&q=reinforcement+learning+feedforward&hl=en&as_sdt=0,5&as_vis=1
+ * 
+ */
+
+namespace Shin
+{
+namespace NeuronNetwork
+{
+namespace Learning
+{
+	class QLearning
+	{
+		public:
+			QLearning(size_t input, size_t size, size_t choices);
+			~QLearning();
+
+			QLearning(const QLearning&) =delete;
+			QLearning& operator=(const QLearning&) =delete;
+
+			void learnDelayed(std::vector<std::pair<Solution,Problem>> &p, float quality);
+			void learnDelayed(std::vector<std::pair<Problem,int>> &p, float quality);
+			void learn(Solution &s, Problem &p, float quality);
+			void learn(Problem &p,int action, float quality);
+
+			void learnNetwork(double maxError=0.01);
+			void learningCoeficient(double t);
+
+			void initialise(size_t input, size_t size,size_t choices);
+			int getChoice(Problem &p);
+			Solution getSolution(Problem &p) {return fun.getSolution(p);}
+			void setLearningCoeficient(double ok, double err) {fun.setLearningCoeficient(ok,err);};
+			void opticalBackPropagation() {fun.opticalBackPropagation();};
+		protected:
+			RL::QFunctionNetwork fun;
+	};
+}
+}
+}
+#endif

src/NeuronNetwork/Learning/RL/QFunction.cpp

@@ -1,6 +1,174 @@
 #include "./QFunction"
 
-Shin::NN::RL::QFunction::QFunction()
+Shin::NeuronNetwork::RL::QFunction::QFunction():learningCoeficient(0.1)
 {
 
 }
+
+Shin::NeuronNetwork::RL::QFunction::~QFunction()
+{
+
+}
+
+Shin::NeuronNetwork::RL::QFunctionTable::QFunctionTable(float maxIODiff) :QFunction(),diff(maxIODiff),data(Shin::NeuronNetwork::RL::QFunctionTable::IOComparator(maxIODiff))
+{
+
+}
+
+float Shin::NeuronNetwork::RL::QFunctionTable::getScore(Shin::NeuronNetwork::Solution& s, Shin::NeuronNetwork::Problem& p)
+{
+	auto element=data.find(p);
+	if(element==data.end())
+		return 0;
+	auto solution=element->second.find(s);
+	if(solution==element->second.end())
+		return 0;
+	return solution->second.first/solution->second.second;
+}
+
+void Shin::NeuronNetwork::RL::QFunctionTable::learn(Shin::NeuronNetwork::Solution& s, Shin::NeuronNetwork::Problem& p, float quality)
+{
+	auto element=data.find(p);
+	if(element==data.end())
+	{
+		data[p]=std::map<Solution,std::pair<float,int>,Shin::NeuronNetwork::RL::QFunctionTable::IOComparator>(Shin::NeuronNetwork::RL::QFunctionTable::IOComparator(diff));
+		element=data.find(p);
+	}
+	auto solution=element->second.find(s);
+	if(solution==element->second.end())
+	{
+		element->second[s].first=quality;
+		element->second[s].second=1;
+	}else
+	{
+		solution->second.first+=quality;
+		solution->second.second++;
+	}
+}
+
+void Shin::NeuronNetwork::RL::QFunctionTable::learnDelayed(std::vector< std::pair< Shin::NeuronNetwork::Solution, Shin::NeuronNetwork::Problem > >& p, float quality)
+{
+	for(int i=p.size()-1;i>=0;i--)
+	{
+		auto &pair=p[i];
+		learn(pair.first,pair.second,quality);
+		quality*=0.3;
+	}
+}
+
+
+Shin::NeuronNetwork::RL::QFunctionNetwork::QFunctionNetwork() : b(nullptr),function(nullptr)
+{
+
+}
+
+Shin::NeuronNetwork::RL::QFunctionNetwork::~QFunctionNetwork()
+{
+	if(function !=nullptr)
+	{
+		delete b;
+		delete function;
+	}
+}
+
+void Shin::NeuronNetwork::RL::QFunctionNetwork::initialiseNetwork(size_t input, size_t size,size_t choices)
+{
+	if(function == nullptr)
+	{
+		function = new FeedForwardNetworkQuick({(int)input,(int)size,(int)choices});
+		b= new Learning::BackPropagation(*function);
+	}
+}
+
+void Shin::NeuronNetwork::RL::QFunctionNetwork::learnDelayed(std::vector< std::pair< Shin::NeuronNetwork::Solution, Shin::NeuronNetwork::Problem > >& p, float quality)
+{
+	if(quality>0)
+	{
+		b->setLearningCoeficient(learningA);
+//		b->setLearningCoeficient(0.05);
+	}else
+	{
+		b->setLearningCoeficient(learningB);
+//		b->setLearningCoeficient(0.008);
+	}
+	
+	for(int i=p.size()-1;i>=0;i--)
+	{
+		learn(p[i].first,p[i].second,quality);
+		quality*=0.95;
+	}
+}
+
+void Shin::NeuronNetwork::RL::QFunctionNetwork::learnDelayed(std::vector< std::pair< Shin::NeuronNetwork::Problem,int> >& p, float quality) // TODO there must be better way
+{
+	std::vector<std::pair<Solution,Problem>> q;
+	register int solSize=0;
+	if(p.size()>0)
+		solSize=function->solve(p[0].first).size();
+	if (!solSize)
+		return;
+	for(int i=0;i<p.size();i++)
+	{
+		Solution s;
+		for(int j=0;j<solSize;j++)
+		{
+			s.push_back(j==p[i].second?1:0);
+		}
+		q.push_back(std::pair<Solution,Problem>(s,p[i].first));
+	}
+	learnDelayed(q,quality);
+}
+
+
+void Shin::NeuronNetwork::RL::QFunctionNetwork::learn(Shin::NeuronNetwork::Solution& s, Shin::NeuronNetwork::Problem& p, float quality)
+{
+	register int loops=abs(quality)/10;
+	if(quality > 0)
+	{
+		while(loops-- >= 0)
+		{
+			b->teach(p,s);
+		}
+	}else
+	{
+		Shin::NeuronNetwork::Solution e;
+		for(int j=s.size()-1;j>=0;j--)
+		{
+			e.push_back(((float)(100-(rand()%101)))/100.0);
+		}
+		while(loops-->=0)
+		{
+			b->teach(p,e);
+		}
+	}
+}
+
+void Shin::NeuronNetwork::RL::QFunctionNetwork::learn(Shin::NeuronNetwork::Problem& p, int choice, float quality)
+{
+	if(quality>0)
+	{
+		b->setLearningCoeficient(learningA);
+	}else
+	{
+		b->setLearningCoeficient(learningB);
+	}
+//	int choice= getChoice(p);
+	Shin::NeuronNetwork::Solution s=function->solve(p);
+	learn(s,p,quality);
+}
+
+int Shin::NeuronNetwork::RL::QFunctionNetwork::getChoice(Shin::NeuronNetwork::Problem& p)
+{
+	Solution s=function->solve(p);
+	float max=-888888888888; //TODO
+	int j=0;
+	for(size_t i=0;i<s.size();i++)
+	{
+		if(s[i]>max)
+		{
+			max=s[i];
+			j=i;
+		}
+	}
+	return j;
+}

src/NeuronNetwork/Learning/RL/QFunction.h

@@ -1,6 +1,14 @@
+#ifndef _Q_FUNCTION_H_
+#define _Q_FUNCTION_H_
+
+#include "../../Solution"
+#include "../../FeedForwardQuick"
+#include "../BackPropagation.h"
+#include "../OpticalBackPropagation.h"
+#include <map>
 namespace Shin
 {
-namespace NN
+namespace NeuronNetwork
 {
 namespace RL
 {
@@ -8,11 +16,81 @@ class QFunction
 	{
 		public:
 			QFunction();
+			virtual ~QFunction();
+			virtual void learnDelayed(std::vector<std::pair<Solution,Problem>> &p, float quality)=0;
+			virtual void learn(Solution &s, Problem &p, float quality)=0;
 		protected:
+			float learningCoeficient;
+	};
 
+	class QFunctionTable : public QFunction
+	{
+		public:
+			QFunctionTable(float maxIODiff=0.01);
+			virtual void learnDelayed(std::vector<std::pair<Solution,Problem>> &p, float quality) override;
+			virtual void learn(Solution &s, Problem &p, float quality) override;
+			virtual float getScore(Solution &s, Problem &p);
+			inline size_t size() {return data.size();}
+			auto begin() {return data.begin();};
+			auto end() {return data.end();};
+		protected:
+			float diff;
+			struct IOComparator
+			{
+				IOComparator(float maxDiff=0):maxIODifference(maxDiff) {}
+				bool operator() (const IO& l, const IO& r) const
+				{
+					if(l.size()!=r.size())
+					{
+						if(l.size() <r.size())
+							return true;
+						return false;
+					}
+					for(size_t i=0;i<l.size();i++)
+					{
+						register float diff=l[i]-r[i];
+						if(fabs(diff) > maxIODifference)
+						{
+							if(diff < maxIODifference)
+								return false;
+							else
+								return true;
+						}
+					}
+					return false;
+				}
+				float maxIODifference=0.1;
+			};
+			std::map<Problem,std::map<Solution,std::pair<float,int>,IOComparator>,IOComparator> data;
+	};
+
+	class QFunctionNetwork : public QFunction
+	{
+		public:
+			QFunctionNetwork();
+			QFunctionNetwork(const QFunctionNetwork&) =delete;
+			virtual ~QFunctionNetwork();
+			
+			QFunctionNetwork operator=(const QFunctionNetwork&) =delete;
+			void initialiseNetwork(size_t input, size_t size,size_t choices);
+
+			virtual void learnDelayed(std::vector<std::pair<Solution,Problem>> &p, float quality) override;
+			virtual void learnDelayed(std::vector<std::pair<Problem,int>> &p, float quality);
+			virtual void learn(Solution &s, Problem &p, float quality) override;
+			virtual void learn(Problem &s, int choice, float quality);
+
+			virtual int getChoice(Problem &p);
+			virtual Solution getSolution(Problem &p) {return function->solve(p);}
+			void setLearningCoeficient(double ok, double err) {learningA=ok;learningB=err;};
+			void opticalBackPropagation() {delete b; b=new Learning::OpticalBackPropagation(*function);};
 		private:
-		
+			Learning::BackPropagation *b;
+			FeedForwardNetworkQuick * function;
+			float learningA=0.05;
+			float learningB=0.008;
 	};
 }
 }
 }
+
+#endif

src/NeuronNetwork/Learning/Reinforcement.cpp

@@ -20,20 +20,22 @@ float Shin::NeuronNetwork::Learning::Reinforcement::learn(const Shin::NeuronNetw
 	//network[2]->operator[](0)->setWeight(0,-5);
 	Solution s=network.solve(problem);
 	float quality=qualityFunction(problem,s);
-	std::vector<float> q;
+	Solution q;
 	//std::cerr << s[0] << "\n";
+
+	register int i=abs((int)quality);
+
+	if(quality > 0)
+	{
 		for(register size_t j=0;j<s.size();j++)
 		{
 			q.push_back(s[j]);//*((float)(990+(rand()%21))/1000.0));
 		}
-
+	}else if(quality <= 0 && i >= 0)
-	register int i=abs((int)quality);
-
-	if(quality <= 0 && i > 0)
 	{
 		for(register size_t j=0;j<s.size();j++)
 		{
-			q[j]=((float)(100-(rand()%101)))/100.0;
+			q.push_back(((float)(100-(rand()%101)))/100.0);
 		}
 	}
 

src/NeuronNetwork/Learning/Unsupervised.h

@@ -1,5 +1,5 @@
 #ifndef _UNSUPERVISEDLEARNING_H_
-#define _UNSUPERVIESDLERANING_H_
+#define _UNSUPERVISEDLEARNING_H_
 
 #include <math.h>
 #include <cstddef>

src/NeuronNetwork/Makefile

@@ -1,6 +1,6 @@
 OBJFILES= Neuron.o Network.o FeedForward.o FeedForwardQuick.o\
 	Learning/Supervised.o Learning/BackPropagation.o Learning/OpticalBackPropagation.o\
-	Learning/Unsupervised.o Learning/Reinforcement.o\
+	Learning/Unsupervised.o Learning/Reinforcement.o Learning/RL/QFunction.o Learning/QLearning.o\
 	Solution.o Problem.o ./IO.o
 
 LINKFILES= ../sse_mathfun.o

src/NeuronNetwork/Network.cpp

@@ -21,7 +21,7 @@ Layer::~Layer()
 
 Solution Layer::solve(const std::vector<float> &input)
 {
-	std::vector <float> ret;
+	Solution ret;
 	for(Neuron *n:neurons)
 	{
 		ret.push_back(n->output(input));

src/NeuronNetwork/Problem.cpp

@@ -7,7 +7,7 @@ Problem::Problem()
 
 }
 
-Problem::operator std::vector<float>() const
+Problem::Problem(std::vector< float >& p): IO(p)
 {
-	return representation();
+
 }

src/NeuronNetwork/Problem.h

@@ -3,18 +3,17 @@
 
 #include <cstddef>
 #include <vector>
+#include "IO.h"
 
 namespace Shin
 {
 namespace NeuronNetwork
 {
-	class Problem
+	class Problem : public IO
 	{
 		public:
 			Problem();
-			virtual ~Problem(){};
+			Problem(std::vector<float> &p);
-			operator std::vector<float>() const;
-			virtual std::vector<float> representation() const =0;
 		protected:
 		private:
 	};

src/NeuronNetwork/Solution.cpp

@@ -6,3 +6,13 @@ Solution::Solution(std::vector<float>&sol):IO(sol)
 {
 
 }
+
+Solution::Solution(std::vector< float > solution):IO(solution)
+{
+
+}
+
+void Solution::push_back(float d)
+{
+	data.push_back(d);
+}

src/NeuronNetwork/Solution.h

@@ -11,8 +11,11 @@ namespace NeuronNetwork
 	class Solution : public IO
 	{
 		public:
-			Solution(const Problem& p);
+			Solution(): IO() {};
+			Solution(const Problem& p) :IO(p) {};
 			Solution(std::vector<float> &solution);
+			Solution(std::vector<float> solution);
+			void push_back(float);
 	};
 }
 }

tests/Makefile

@@ -7,8 +7,10 @@ NN_TESTS= \
 	nn-01 nn-02 nn-03 nn-bp-sppeed \
 	nn-bp-xor \
 	nn-obp-xor \
-	nn-rl-xor nn-rl-and \
+	nn-rl-xor nn-rl-and nn-rl-qfun\
 	nn-reinforcement nn-04
+#	nn-test nn-rl-qfun\
+
 ALL_TESTS=$(NN_TESTS) $(GEN_TESTS)
 
 LIBS=$(LIB_DIR)/Genetics.a $(LIB_DIR)/NeuronNetwork.a

tests/nn-01.cpp

@@ -5,20 +5,15 @@
 #include <iostream>
 #include <vector>
 
+//typedef Shin::NeuronNetwork::Problem X;
+
 class X: public Shin::NeuronNetwork::Problem
 {
 	public:
-		X(const X& a) :q(a.q) {}
+		X(const X& a) :Problem(a) {}
-		X(const std::vector<float> &a):q(a) {}
+		X(const std::vector<bool> &a):Problem() { for (bool s:a) data.push_back((float)s);}
-		X(const std::vector<bool> &a):q() {for(bool s:a) q.push_back((float)s);}
-		std::vector<float> representation() const
-		{
-			return q;
-		}
 	protected:
-		std::vector<float> q;
 };
-
 int main(int argc,char**)
 {
 	srand(time(NULL));

tests/nn-03.cpp

@@ -8,7 +8,7 @@
 class X: public Shin::NeuronNetwork::Problem
 {
 	public:
-		X(const X& a) :q(a.q) {}
+		X(const X& a) :Problem(),q(a.q) {}
 		X(const std::vector<float> &a):q(a) {}
 		std::vector<float> representation() const
 		{
@@ -33,8 +33,9 @@ int main()
 	s.push_back(Shin::NeuronNetwork::Solution(std::vector<float>({1})));
 	p.push_back(X(std::vector<float>({1,1})));
 
-	Shin::NeuronNetwork::FeedForwardNetworkQuick q({2,2,4,1});
+	Shin::NeuronNetwork::FeedForwardNetworkQuick q({2,4,1});
 	Shin::NeuronNetwork::Learning::BackPropagation b(q);
+	b.setLearningCoeficient(10);
 
 	b.debugOn();
 	for(int i=0;i<4;i++)
@@ -45,7 +46,7 @@ int main()
 	}
 	b.debugOff();
 
-	for(int i=0;i<4000;i++)
+	for(int i=0;i<40000;i++)
 	{
 		b.teach(p[i%4],s[i%4]);
 	}

tests/nn-bp-xor.cpp

@@ -7,14 +7,8 @@
 class X: public Shin::NeuronNetwork::Problem
 {
 	public:
-		X(const X& a) :q(a.q) {}
+		X(const X& a) :Problem(a) {}
-		X(const std::vector<float> &a):q(a) {}
+		X(const std::vector<float> &a):Problem() {data=a;}
-		std::vector<float> representation() const
-		{
-			return q;
-		}
-	protected:
-		std::vector<float> q;
 };
 
 int main()
@@ -66,7 +60,7 @@ int main()
 				std::cerr << j << "(" << err <<"):\n";
 				for(int i=0;i<4;i++)
 				{
-					std::cerr << "\t" << i%4 <<". FOR: [" << p[i%4]->representation()[0] << "," <<p[i%4]->representation()[1] << "] res: " <<
+					std::cerr << "\t" << i%4 <<". FOR: [" << p[i%4]->operator[](0) << "," <<p[i%4]->operator[](1) << "] res: " <<
 							q.solve(*p[i%4])[0] << " should be " << s[i%4]->operator[](0)<<"\n";
 				}
 			}

tests/nn-obp-xor.cpp

@@ -7,14 +7,8 @@
 class X: public Shin::NeuronNetwork::Problem
 {
 	public:
-		X(const X& a) :q(a.q) {}
+		X(const X& a) :Problem(a) {}
-		X(const std::vector<float> &a):q(a) {}
+		X(const std::vector<float> &a):Problem() {data=a;}
-		std::vector<float> representation() const
-		{
-			return q;
-		}
-	protected:
-		std::vector<float> q;
 };
 
 int main()
@@ -67,7 +61,7 @@ int main()
 				std::cerr << j << "(" << err <<"):\n";
 				for(int i=0;i<4;i++)
 				{
-					std::cerr << "\t" << i%4 <<". FOR: [" << p[i%4]->representation()[0] << "," <<p[i%4]->representation()[1] << "] res: " <<
+					std::cerr << "\t" << i%4 <<". FOR: [" << p[i%4]->operator[](0) << "," <<p[i%4]->operator[](1) << "] res: " <<
 							q.solve(*p[i%4])[0] << " should be " << s[i%4]->operator[](0)<<"\n";
 				}
 			}

tests/nn-rl-and.cpp

@@ -8,14 +8,8 @@
 class X: public Shin::NeuronNetwork::Problem
 {
 	public:
-		X(const X& a) :q(a.q) {}
+		X(const X& a) :Problem(a) {}
-		X(const std::vector<float> &a):q(a) {}
+		X(const std::vector<float> &a):Problem() {data=a;}
-		std::vector<float> representation() const
-		{
-			return q;
-		}
-	protected:
-		std::vector<float> q;
 };
 
 int main()
@@ -28,14 +22,17 @@ int main()
 
 	p.push_back(new X(std::vector<float>({1,1})));
 
-	Shin::NeuronNetwork::FeedForwardNetworkQuick q({1,1});
+	p.push_back(new X(std::vector<float>({1,0})));
+	p.push_back(new X(std::vector<float>({0,1})));
+
+	Shin::NeuronNetwork::FeedForwardNetworkQuick q({2,1});
 	Shin::NeuronNetwork::Learning::Reinforcement b(q);
 	int i=0;
-	double targetQuality=1.4;
+	double targetQuality=0.5;
 	b.setQualityFunction(
 		[](const Shin::NeuronNetwork::Problem &pr,const Shin::NeuronNetwork::Solution &s)->float
 		{
-			if(pr.representation()[0]==0)
+			if(pr[0]==1 && pr[1]==1)
 			{
 				//ocekavame 1
 				int e=(s[0]-0.80)*15.0;//+(abs(s[1])-0.5)*100.0;
@@ -54,12 +51,12 @@ int main()
 	
 		if(i%100000==0)
 			srand(time(NULL));
-		if(err > targetQuality)
+		if(err > targetQuality||i%1000==0)
 		{
 			std::cerr << i << " ("<< err <<").\n";
-			for(int j=0;j<2;j++)
+			for(int j=0;j<4;j++)
 			{
-				std::cerr << j%4 <<". FOR: [" << p[j%4]->representation()[0] << "," <<p[j%4]->representation()[0] << "] res: " << q.solve(*p[j%4])[0] << "\n";
+				std::cerr << j%4 <<". FOR: [" << p[j%4]->operator[](0) << "," <<p[j%4]->operator[](0) << "] res: " << q.solve(*p[j%4])[0] << "\n";
 			}
 		}
 		if(err >targetQuality)

tests/nn-rl-xor.cpp

@@ -5,19 +5,15 @@
 #include <iostream>
 #include <vector>
 
+
 class X: public Shin::NeuronNetwork::Problem
 {
 	public:
-		X(const X& a) :q(a.q) {}
+		X(const X& a) :Problem(a) {}
-		X(const std::vector<float> &a):q(a) {}
+		X(const std::vector<float> &a):Problem() {data=a;}
-		std::vector<float> representation() const
-		{
-			return q;
-		}
-	protected:
-		std::vector<float> q;
 };
 
+
 int main()
 {
 	srand(time(NULL));
@@ -26,9 +22,9 @@ int main()
 		Shin::NeuronNetwork::FeedForwardNetworkQuick q({2,4,1});
 		Shin::NeuronNetwork::Learning::Reinforcement b(q);
 		//b.setPropagator(new Shin::NeuronNetwork::Learning::OpticalBackPropagation(q));
-		b.getPropagator().setLearningCoeficient(3);
+		b.getPropagator().setLearningCoeficient(0.4);
 		//b.getPropagator().allowEntropy();
-		double targetQuality =1;
+		double targetQuality =2.9;
 		if(test==2)
 		{
 			targetQuality =1.62;
@@ -38,9 +34,8 @@ int main()
 			b.getPropagator().setLearningCoeficient(0.5);
 		}
 		b.setQualityFunction(
-			[](const Shin::NeuronNetwork::Problem &pr,const Shin::NeuronNetwork::Solution &s)->float
+			[](const Shin::NeuronNetwork::Problem &p,const Shin::NeuronNetwork::Solution &s)->float
 			{
-				std::vector <float> p=pr;
 				float expect=0.0;
 				if(p[0] && p[1])
 					expect=0;
@@ -55,10 +50,10 @@ int main()
 
 				if(expect==0)
 				{
-					expect=0.1-abs(s[0]);
+					expect=0.3-abs(s[0]);
 				}else
 				{
-					expect=s[0]-0.9;
+					expect=s[0]-0.7;
 				}
 
 //				std::cerr << " returnning " << expect*5.0 << "\n";
@@ -93,7 +88,7 @@ int main()
 				std::cerr << i << " ("<< err <<").\n";
 				for(int j=0;j<4;j++)
 				{
-					std::cerr << "\t" << j%4 << ". FOR: [" << p[j%4]->representation()[0] << "," <<p[j%4]->representation()[1] << "] res: " <<
+					std::cerr << "\t" << i%4 <<". FOR: [" << p[j%4]->operator[](0) << "," <<p[j%4]->operator[](1) << "] res: " <<
 							q.solve(*p[j%4])[0] << "\n";
 				}
 			}