// Get normal model of specific learning-method, learning will abort if shouldContinueLearning() == false and will return null;
public static List <CorrelatedFeatures> GetNormal(Dictionary <string, List <float> > features, string method, ShouldContinue shouldContinueLearning, bool commutative = true)
{
if (!ThresholdMethods.ContainsKey(method))
{
return(new List <CorrelatedFeatures>());
}
return(GetNormal(features, ThresholdMethods[method], shouldContinueLearning, commutative));
}
} // transformation-specific data (describe the transformation)
public AnimationItem(AnimationType type = AnimationType.None, ShouldContinue condition = null, float[] data = null)
{
Type = type;
Data = data;
IsStarted = false;
}
// Get normal model of specific learning-method, learning will abort if shouldContinueLearning() == false and will return null;
private static List <CorrelatedFeatures> GetNormal(Dictionary <string, List <float> > features, ThresholdFactory create, ShouldContinue shouldContinueLearning, bool commutative = true)
{
List <CorrelatedFeatures> result = new List <CorrelatedFeatures>();
List <String> orderedFeatures = new List <string>();
foreach (var s in features.Keys)
{
orderedFeatures.Add(s);
}
// for each feature as being feautre 1 [left feature]
for (int i = 0; i < orderedFeatures.Count; i++)
{
// check if it's still relevant to learn[ or model has already deleted, for example]
// check only in outer loop, but inner loop is very fast
if (!shouldContinueLearning())
{
return(null);
}
int mostCorrelative_idx = -1;
float mostCorrelative_pearson = 0;
string f1 = orderedFeatures[i];
// for each another feature 2 as being right feature, find the most correlative feature 2 to feature 1 (larger abs(pearson))
// commutative-param: set if NOT to use upper triangle matrix method to find correlation,
// if commutative-param and A-B correlation then B-A correlation (since correlation dependes on most abs pearson which is commutative-function)
int j = commutative ? 0 : i + 1;
for (; j < orderedFeatures.Count; j++)
{
if (i == j)
{
continue;
}
string f2 = orderedFeatures[j];
float[] x = features[f1].ToArray();
float[] y = features[f2].ToArray();
var p = MathUtil.Pearson(x, y);
if (Math.Abs(p) <= Math.Abs(mostCorrelative_pearson))
{
continue;
}
mostCorrelative_idx = j;
mostCorrelative_pearson = p;
}
if (mostCorrelative_idx == -1)
{
continue;
}
// partial setup for <float threshold, string typeName, T correlationObject_T> fields,
CorrelatedFeatures tmp = create(features[f1].ToArray(), features[orderedFeatures[mostCorrelative_idx]].ToArray());
// or tmp == null if no correlation or error
if (tmp == null)
{
continue;
}
// give 'safe' space
tmp.threshold = tmp.threshold * 1.1f;
if (!IsRegularNum(mostCorrelative_pearson) || !IsRegularNum(tmp.threshold))
{
continue;
}
// set the rest of fields that need to be set-up
tmp.correlation = mostCorrelative_pearson;
tmp.feature1 = f1;
tmp.feature2 = orderedFeatures[mostCorrelative_idx];
result.Add(tmp);
}
if (!shouldContinueLearning())
{
return(null);
}
return(result);
}
