private static float[] GenerateWeights(SportSettings settings, List <Result> trainingCases, Func <Result, bool> expectedResultGetter,
float acceptableError, float alpha, int maxIterations, float initialBias, out float bias)
{
float[] weights = new float[3]; // Age, Height, IMC
int testCount = trainingCases.Count;
float currentBias = initialBias;
float totalError = float.MaxValue;
int elapsedIterations = 0;
// To avoid an infinite loop, either a loop count hard-cap or acceptable error threshold will be reached.
while (elapsedIterations < maxIterations && totalError > acceptableError)
{
for (int i = 0; i < testCount; ++i)
{
Result testCase = trainingCases[i];
bool expectedResult = expectedResultGetter(testCase);
bool output = GetResult(GetCriteriaResults(settings, testCase), weights, currentBias);
int delta = (expectedResult ? 1 : 0) - (output ? 1 : 0);
weights = UpdateWeights(weights, alpha, delta, settings, testCase);
currentBias += alpha * delta;
}
totalError = GetError(settings, trainingCases, expectedResultGetter, weights, currentBias);
elapsedIterations++;
}
bias = currentBias;
return(weights);
}
/// <summary>
/// Obtains an array that describes what sets of criteria a data set has "cleared", and which they haven't.
/// </summary>
public static bool[] GetCriteriaResults(SportSettings settings, IEvaluatable toEvaluate)
{
return(new bool[]
{
(settings.MinAge <= toEvaluate.Edad && toEvaluate.Edad <= settings.MaxAge),
(settings.MinHeight <= toEvaluate.Estatura && toEvaluate.Estatura <= settings.MaxHeight),
(settings.MinIMC <= toEvaluate.IMC && toEvaluate.IMC <= settings.MaxIMC)
});
}
public static float GetError(SportSettings settings, List <Result> trainingCases, Func <Result, bool> expectedResultGetter, float[] currentWeights, float bias)
{
float errorSum = 0;
int size = trainingCases.Count;
for (int i = 0; i < size; ++i)
{
Result testCase = trainingCases[i];
bool expectedResult = expectedResultGetter(testCase);
bool output = GetResult(GetCriteriaResults(settings, testCase), currentWeights, bias);
errorSum += Math.Abs((expectedResult ? 1 : 0) - (output ? 1 : 0));
}
return(errorSum / ((float)size));
}
public static bool[] ExecutePerceptron(int maxIterations, float acceptableError, float alpha, float initialBias,
SportSettings settings, List <Result> trainingCases, Func <Result, bool> expectedResultGetter, List <InputData> toEvaluate, TextBox resultsTextBox)
{
float[] weights = GenerateWeights(settings, trainingCases, expectedResultGetter, acceptableError, alpha, maxIterations, initialBias, out float bias);
float weightsError = GetError(settings, trainingCases, expectedResultGetter, weights, bias);
int toEvaluateSize = toEvaluate.Count;
bool[] results = new bool[toEvaluateSize];
for (int i = 0; i < toEvaluateSize; i++)
{
InputData next = toEvaluate[i];
results[i] = GetResult(GetCriteriaResults(settings, next), weights, bias);
}
resultsTextBox.Text += $"La influencia de cada una de las características físicas para determinar la aptitud fue la siguiente:\nEdad: {weights[0]*100}% | Estatura: {weights[1]*100}% | IMC: {weights[2]*100}%\n";
resultsTextBox.Text += $"El error final para estos pesos fue de {weightsError * 100}%.\n\n";
return(results);
}
public static float[] UpdateWeights(float[] currentWeights, float alpha, float delta, SportSettings settings, IEvaluatable toEvaluate)
{
bool[] criteriaResults = GetCriteriaResults(settings, toEvaluate);
float[] result = new float[3]; // Age, Height, IMC
int size = result.Length;
for (int i = 0; i < size; i++)
{
result[i] = currentWeights[i] + (alpha * delta * (criteriaResults[i] ? 1 : 0));
}
return(result);
}