public double calculate_discrimenet_function(Generic_State_Of_Nature object_state_of_nature, double[,] vector_observed_features_values)
{
double gi_x = 0;
Matrix<double> X = Matrix.Build.DenseOfArray(vector_observed_features_values);
/**
* todo move this code to training step
**/
Matrix<double> Wi = (-1 / 2) * object_state_of_nature.Matrix_Sigma_Inverse;
Matrix<double> wi = object_state_of_nature.Matrix_Sigma_Inverse * object_state_of_nature.Vector_Meu;
Matrix<double> omegai_0 = (-1 / 2) * object_state_of_nature.Vector_Meu_Transpose * object_state_of_nature.Matrix_Sigma_Inverse * object_state_of_nature.Vector_Meu - (1 / 2) * Math.Log(object_state_of_nature.Matrix_Sigma.Determinant(),Math.E) + Math.Log(object_state_of_nature.priori, Math.E);
Matrix<double> ret = X.Transpose() * Wi * X + wi.Transpose() * X + omegai_0;
gi_x = ret.Determinant();
return gi_x;
}
public GenericDataSet(
Stream _data_set_file_stream,
char _file_delimeter,
int _number_of_features,
int _number_of_states_of_nature,
int _number_of_samples_per_state_of_nature,
int _number_of_training_samples_per_state_of_nature,
int _number_of_test_samples_per_state_of_nature)
{
number_of_features = _number_of_features;
number_of_states_of_nature = _number_of_states_of_nature;
number_of_samples_per_state_of_nature = _number_of_samples_per_state_of_nature;
number_of_training_samples_per_state_of_nature = _number_of_training_samples_per_state_of_nature;
number_of_test_samples_per_state_of_nature = _number_of_test_samples_per_state_of_nature;
double priori = 1 / number_of_states_of_nature; // @note: equal priori results in uniform decision boundary mainly determined by the posteriori
array_of_states_natures = new Generic_State_Of_Nature[number_of_states_of_nature];
for (int i = 0; i < number_of_states_of_nature; i++)
{
array_of_states_natures[i] = new Generic_State_Of_Nature(
number_of_features,
number_of_training_samples_per_state_of_nature,
number_of_test_samples_per_state_of_nature,
priori
);
}
read_data_set_from_file_stream(_data_set_file_stream, _file_delimeter);
for (int i = 0; i < number_of_states_of_nature; i++)
{
array_of_states_natures[i].calculate_meus_vector_from_samples();
array_of_states_natures[i].calculate_covariance_matrix_from_samples();
}
}
public int classify_using_discriminent_function(Generic_State_Of_Nature[] array_generic_state_nature, double[,] vector_observed_features_values)
{
double maximum_discriminent_value = double.PositiveInfinity;
int class_index = -1;
for (int i = 0; i < array_generic_state_nature.Length; i++)
{
double discriminent_value = calculate_discrimenet_function(array_generic_state_nature[i], vector_observed_features_values);
if (discriminent_value > maximum_discriminent_value)
{
maximum_discriminent_value = discriminent_value;
class_index = i;
}
}
return class_index;
}
