refactor models (#636)

Signed-off-by: Andrey Parfenov <a1994ndrey@gmail.com>

cpp_package/src/inc/data_filter.h

@@ -193,7 +193,8 @@ class DataFilter
      * @return oxygen level
      */
     static double get_oxygen_level (double *ppg_ir, double *ppg_red, int data_len,
-        int sampling_rate, double coef1 = 0.0, double coef2 = -37.663, double coef3 = 114.91);
+        int sampling_rate, double coef1 = 1.5958422, double coef2 = -34.6596622,
+        double coef3 = 112.6898759);
     /**
      * calculate heart rate
      * @param ppg_ir input 1d array

csharp_package/brainflow/brainflow/data_filter.cs

@@ -306,7 +306,7 @@ public static double get_railed_percentage (double[] data, int gain)
         /// <param name="coef2">approximation coef</param>
         /// /// <param name="coef3">intercept for approximation</param>
         /// <returns>stddev</returns>
-        public static double get_oxygen_level (double[] ppg_ir, double[] ppg_red, int sampling_rate, double coef1 = 0.0, double coef2 = -37.663, double coef3 = 114.91)
+        public static double get_oxygen_level (double[] ppg_ir, double[] ppg_red, int sampling_rate, double coef1 = 1.5958422, double coef2 = -34.6596622, double coef3 = 112.6898759)
         {
             if (ppg_ir.Length != ppg_red.Length)
             {

java_package/brainflow/src/main/java/brainflow/DataFilter.java

@@ -251,7 +251,7 @@ public static double get_oxygen_level (double[] ppg_ir, double[] ppg_red, int sa
      */
     public static double get_oxygen_level (double[] ppg_ir, double[] ppg_red, int sampling_rate) throws BrainFlowError
     {
-        return get_oxygen_level (ppg_ir, ppg_red, sampling_rate, 0.0, -37.663, 114.91);
+        return get_oxygen_level (ppg_ir, ppg_red, sampling_rate, 1.5958422, -34.6596622, 112.6898759);
     }
 
     /**

julia_package/brainflow/src/data_filter.jl

@@ -253,7 +253,7 @@ end
     return output[1]
 end
 
-@brainflow_rethrow function get_oxygen_level(ppg_ir, ppg_red, sampling_rate::Integer, coef1=0.0, coef2=-37.663, coef3=114.91)
+@brainflow_rethrow function get_oxygen_level(ppg_ir, ppg_red, sampling_rate::Integer, coef1=1.5958422, coef2=-34.6596622, coef3=112.6898759)
     if length(ppg_ir) != length(ppg_red)
       throw(BrainFlowError(string("invalid size", INVALID_ARGUMENTS_ERROR), Integer(INVALID_ARGUMENTS_ERROR)))
     end

python_package/brainflow/data_filter.py

@@ -771,7 +771,7 @@ def get_railed_percentage(cls, data: NDArray[Float64], gain: int):
 
     @classmethod
     def get_oxygen_level(cls, ppg_ir: NDArray[Float64], ppg_red: NDArray[Float64], sampling_rate: int,
-                         coef1=0.0, coef2=-37.663, coef3=114.91):
+                         coef1=1.5958422, coef2=-34.6596622, coef3=112.6898759):
         """get oxygen level from ppg
 
         :param ppg_ir: input array

src/ml/generated/mindfulness_model.cpp

@@ -1,6 +1,6 @@
 
 #include "mindfulness_model.h"
 // clang-format off
-const double mindfulness_coefficients[5] = {2.6338144674136394,4.006742906593334,-34.51389221061297,1.1950604401540308,35.78022137767881};
-double mindfulness_intercept = 0.364078;
+const double mindfulness_coefficients[5] = {-1.4765769283767163,2.6620930328900974,-31.43942997194057,9.464066586727622,45.512106420941684};
+double mindfulness_intercept = 0.000000;
 // clang-format on

src/ml/train/train_classifiers.py

@@ -42,6 +42,8 @@ def prepare_data(first_class, second_class, blacklisted_channels=None):
     dataset_y = list()
     for data_type in (first_class, second_class):
         for file in glob.glob(os.path.join('data', data_type, '*', '*.csv')):
+            data_x_temp = list()
+            data_y_temp = list()
             logging.info(file)
             board_id = os.path.basename(os.path.dirname(file))
             try:
@@ -58,13 +60,17 @@ def prepare_data(first_class, second_class, blacklisted_channels=None):
                         feature_vector = bands[0]
                         feature_vector = feature_vector.astype(float)
                         dataset_x.append(feature_vector)
+                        data_x_temp.append(feature_vector)
                         if data_type == first_class:
                             dataset_y.append(0)
+                            data_y_temp.append(0)
                         else:
                             dataset_y.append(1)
+                            data_y_temp.append(0)
                         cur_pos = cur_pos + int(window_size * overlaps[num] * sampling_rate)
             except Exception as e:
                 logging.error(str(e), exc_info=True)
+            print_dataset_info((data_x_temp, data_y_temp))
 
     logging.info('1st Class: %d 2nd Class: %d' % (len([x for x in dataset_y if x == 0]), len([x for x in dataset_y if x == 1])))
 
@@ -115,7 +121,7 @@ def print_dataset_info(data):
 
 def train_regression_mindfulness(data):
     model = LogisticRegression(solver='liblinear', max_iter=4000,
-                                penalty='l2', random_state=2, fit_intercept=True, intercept_scaling=0.2)
+                                penalty='l2', random_state=2, fit_intercept=False, intercept_scaling=3)
     logging.info('#### Logistic Regression ####')
     scores = cross_val_score(model, data[0], data[1], cv=5, scoring='f1_macro', n_jobs=8)
     logging.info('f1 macro %s' % str(scores))
@@ -207,14 +213,14 @@ def main():
             dataset_y = pickle.load(f)
         data = dataset_x, dataset_y
     else:
-        data = prepare_data('relaxed', 'focused', blacklisted_channels={'T3', 'T4'})
+        data = prepare_data('relaxed', 'focused')
     print_dataset_info(data)
     train_regression_mindfulness(data)
-    train_svm_mindfulness(data)
-    train_knn_mindfulness(data)
-    train_random_forest_mindfulness(data)
-    train_mlp_mindfulness(data)
-    train_stacking_classifier(data)
+    #train_svm_mindfulness(data)
+    #train_knn_mindfulness(data)
+    #train_random_forest_mindfulness(data)
+    #train_mlp_mindfulness(data)
+    #train_stacking_classifier(data)
 
 
 if __name__ == '__main__':