/// <summary>
/// The estimation of classifier parameters has to run in two cycles. This is the Preprocess Phrase. In order to later compute
/// the variance of each category/feature combination, the avarage of each category/feature combination has to be computed before.
///
/// The count of items for each category is computed to be used later to obtain the apriory probability.
/// The count of items for each category/feature combination is computed to later obtain the posteriory probability.
/// </summary>
/// <param name="examples"></param>
private void PreprocessData()
{
for (int i = 0; i < _itemsCount; i++)
{
int category = ConvertToCategoryIdentifier(Description.Label.Type, Y[i]);
CategoryItemsCount[category]++;
var values = X[i];
int j = 0;
for (int helper = 0; helper < Description.Features.Count(); helper++)
{
var feature = Description.Features[helper];
if (ContinuesTypes.Contains(feature.Type))
{
var value = values[j];
// add the value to the avarage (later this will be divided by the count to obtain the avarage).
CategoryFeatureAvg[category][j] += value;
//have a small list of values for each feature/category
if (CategoryFeatureValues[category][j] == null)
{
CategoryFeatureValues[category][j] = new List <double>();
}
CategoryFeatureItemsCount[category][j]++;
CategoryFeatureValues[category][j].Add(value);
}
if (feature is StringProperty)
{
var sp = feature as StringProperty;
var wordCount = sp.Dictionary.Count();
//each string is converted into binary vector. I will loop over the vector representing the string.
for (int k = 0; k < wordCount; k++)
{
if (values[j + k] == 1)
{
CategoryFeatureItemsCount[category][j + k]++;
}
}
j += wordCount - 1;
}//string property
if (feature.Type == typeof(bool))
{
if (values[j] == 1)
{
CategoryFeatureItemsCount[category][j]++;
}
}
j++;
} //features
} //items
}
/// <summary>
/// This method builds the model - it fills the Apriory,Posteriory, CategoryFeatureAvg and CategoryFeatureVariance with approriate values.
/// </summary>
private void BuildTheModel()
{
_itemsCount = Y.Count();
_featuresCount = X.Cols;
_categoryCount = Y.Distinct().Count();
//initialize two dimensional arrays
InitializeArrays();
//prepare the data - avg, counts of items foreach category/feature.
PreprocessData();
for (int i = 0; i < _categoryCount; i++)
{
Apriory[i] = CategoryItemsCount[i] / _itemsCount;
int j = 0;
for (int featureIndex = 0; featureIndex < Description.Features.Count(); featureIndex++)
{
var feature = Description.Features[featureIndex];
if (ContinuesTypes.Contains(feature.Type))
{
//values were added so far to this fied
//no divide the result by the count
CategoryFeatureAvg[i][j] /= CategoryFeatureItemsCount[i][j];
var variance = Helper.Variance(CategoryFeatureValues[i][j].ToArray(), CategoryFeatureAvg[i][j]);
CategoryFeatureVariance[i][j] = variance;
}
if (feature is StringProperty)
{
var sp = feature as StringProperty;
var wordCount = sp.Dictionary.Count();
for (int k = 0; k < wordCount; k++)
{
Posteriory[i][j + k] = GetProbability(i, j + k);
}
j += wordCount - 1;
}
if (feature.Type == typeof(bool))
{
Posteriory[i][j] = GetProbability(i, j);
}
j++;
}
}
}
