/// <summary>
/// This method accepts the test data as input and determines the output for each sample of test data
/// </summary>
/// <param name="data">test data inputs</param>
/// <param name="ctx">test data descriptions</param>
/// <returns>double[]</returns>
public override IResult Predict(double[][] data, IContext ctx)
{
double[][] calcuLatedOutput = new double[m_HiddenLayerNeurons.Length + 1][];
double[][] weightedInputs = new double[m_HiddenLayerNeurons.Length + 1][];
MLPerceptronResult result = new MLPerceptronResult()
{
results = new double[data.Length * m_OutputLayerNeurons],
};
for (int i = 0; i < data.Length; i++)
{
CalcFirstHiddenLayer(data[i], m_InpDims, out calcuLatedOutput[0], out weightedInputs[0]);
CalcRemainingHiddenLayers(calcuLatedOutput[0], weightedInputs[0], m_InpDims, out calcuLatedOutput, out weightedInputs);
CalculateResultatOutputlayer(calcuLatedOutput[m_HiddenLayerNeurons.Length - 1], m_InpDims, m_SoftMax, out calcuLatedOutput[m_HiddenLayerNeurons.Length], out weightedInputs[m_HiddenLayerNeurons.Length]);
for (int j = 0; j < m_OutputLayerNeurons; j++)
{
result.results[m_OutputLayerNeurons * i + j] = calcuLatedOutput[m_HiddenLayerNeurons.Length][j];
}
}
return(result);
}
//[InlineData(new int[] { 6, 3 })]
public void ImageRecognitionTest()
{
int size = 64;
int numberOfOutputs = 1;
int numberOfInputs = size * size;
var context = getImgRecognitionDescriptor(numberOfInputs);
LearningApi api = new LearningApi(context);
api.UseActionModule <object, double[][]>((notUsed, ctx) =>
{
return(getImageData(size, $"{Directory.GetCurrentDirectory()}\\MLPerceptron\\TrainingImages"));
//return getSomOtherData($"{Directory.GetCurrentDirectory()}\\MLPerceptron\\TestFiles\\TrainingData.csv");
});
// High number of hidden neurons in first layer brings network to constant result for everything.
// var hiddenLayerNeurons = new int[] { size*size, 3 };
var hiddenLayerNeurons = new int[] { 6000, 9, 3 };
api.UseMLPerceptron(0.01, 6, 1, 1, hiddenLayerNeurons);
Stopwatch sw = new Stopwatch();
sw.Start();
IScore score = api.Run() as IScore;
sw.Stop();
Trace.WriteLine($"Duration:{(double)(sw.ElapsedMilliseconds / 1000 / 60)} min");
var testImageData = getImageData(size, $"{Directory.GetCurrentDirectory()}\\MLPerceptron\\TestingImages");
//var testImageData= getSomOtherData($"{Directory.GetCurrentDirectory()}\\MLPerceptron\\TestFiles\\TestData.csv");
MLPerceptronResult result = api.Algorithm.Predict(testImageData, api.Context) as MLPerceptronResult;
// float accuracy = MLPHelpers.GetAccuracy(testImageData, result.results, numberOfOutputs);
}
