public void TestMethod1()
{
//number of binary vectors that will be used to train the SP
int numExamples = 10;
//2 dimensional input vector. Each row will be used as one-dimensional vector
int[,] inputVectors = new int[numExamples, 1000];
//2 dimensional output binary vector
int[,] outputColumns = new int[numExamples, 2048];
//Feeding here the input vector with random binary numbers
for (int i = 0; i < numExamples; i++)
{
for (int k = 0; k < 1000; k++)
{
var random = new Random();
inputVectors[i, k] = random.Next(0, 2);
}
}
//Spatial Pooler initialization and configuration
var parameters = Helpers.GetDefaultParams();
parameters.setInputDimensions(new int[] { 1000 });
parameters.setColumnDimensions(new int[] { 2048 });
var sp = new SpatialPooler();
var mem = new Connections();
parameters.apply(mem);
sp.init(mem);
//One dimensional input vector initialized with binary numbers
int[] inputVector = Helpers.GetRandomVector(1000,
parameters.Get <Random>(KEY.RANDOM));
//Array that will be expose to SP to active some of its columns
int[] activeCols = new int[2048];
//List for active columns scores
List <int> activeColScores = new List <int>();
//Learning is turned off
sp.compute(mem, inputVector, activeCols, false);
//overlaps score of the active columns
var overlaps = sp.calculateOverlap(mem, inputVector);
//Sorting reversely overlaps arrays then writing it to the text file overlapBeforeTraining.txt
overlaps = reverseSort(overlaps);
WriteIntArrayToFile("overlapBeforeTraining.txt", overlaps);
//we put the overlap score of each active column in a list called activeColScores
for (int i = 0; i < 2048; i++)
{
if (activeCols[i] != 0)
{
activeColScores.Add(overlaps[i]);
}
}
int numberOfActivCols = 0;
//counting the number of active columns
foreach (int i in activeColScores)
{
numberOfActivCols = numberOfActivCols + 1;
}
// Array to take the number of active columns to write it to file for plotting purpose
int[] numberOfActiveColumns = new int[] { 1 };
numberOfActiveColumns[0] = numberOfActivCols;
int[] inputVectorsRowk = new int[] { };
int[] outputColumnsRowk = new int[] { };
//number of times the vectors are exposed to the SP
int epochs = 1;
//the "numExamples" input binary vectors are exposed to the SP "epochs" times to train it
for (int i = 0; i < epochs; i++)
{
for (int k = 0; k < numExamples; k++)
{
inputVectorsRowk = GetRow(inputVectors, k);
outputColumnsRowk = GetRow(outputColumns, k);
sp.compute(mem, inputVectorsRowk, outputColumnsRowk, true);
}
}
overlaps = sp.calculateOverlap(mem, inputVectorsRowk);
overlaps = reverseSort(overlaps);
WriteIntArrayToFile("overlapAfterTraining.txt", overlaps);
WriteIntArrayToFile("numberOfActCols.txt", numberOfActiveColumns);
//overlap before and after training , numberOfActCols needed for graphs
runPythonCode2("overlapBeforeTraining.txt", "overlapAfterTraining.txt", "numberOfActCols.txt");
int[,] inputVectorsCorrupted = new int[numExamples, 1000];
int[,] outputColumnsCorrupted = new int[numExamples, 2048];
float[] Noise = new float[] { 0, 5, 10, 15, 20, 30, 40, 50, 60, 70, 80, 90, 100 };
double[] percentOverlapInputs = new double[Noise.Length];
double[] percentOverlapOutputs = new double[Noise.Length];
int[] inputVectorsCorruptedRow0 = GetRow(inputVectorsCorrupted, 0);
int[] inputVectorsRow0 = GetRow(inputVectors, 0);
int[] outputColumnsRow0 = GetRow(outputColumns, 0);
int[] outputColumnsCorruptedRow0 = GetRow(outputColumnsCorrupted, 0);
resetVector(inputVectorsRow0, inputVectorsCorruptedRow0);
for (int i = 0; i < Noise.Length; i++)
{
inputVectorsCorruptedRow0 = corruptVector(ref inputVectorsRow0, Noise[i]);
sp.compute(mem, inputVectorsRow0, outputColumnsRow0, false);
sp.compute(mem, inputVectorsCorruptedRow0, outputColumnsCorruptedRow0, false);
percentOverlapInputs[i] = percentOverlap(inputVectorsRow0, inputVectorsCorruptedRow0);
percentOverlapOutputs[i] = percentOverlap(outputColumnsRow0, outputColumnsCorruptedRow0);
}
WriteDoubleArrayToFile("percentOverlapInputs.txt", percentOverlapInputs);
WriteDoubleArrayToFile("percentOverlapOutputs.txt", percentOverlapOutputs);
runPythonCode1("percentOverlapInputs.txt", "percentOverlapOutputs.txt");
}
