Input of the model has been standardized to BWHF (Batch, Width, Height,...

Input of the model has been standardized to BWHF (Batch, Width, Height, Features) shape. RandomScale2DLayer and RandomShift2DLayer has been incorporated into the model in order to apply these two steps of data augmentation on-the-fly during the training of the model, but not during the test of the model.

core/model.py

 import tensorflow as tf
 
-from tensorflow.keras.models import Sequential
-from tensorflow.keras.layers import Dense, Dropout, Flatten, Conv1D, MaxPooling1D
-from tensorflow.keras.layers import LeakyReLU
+from core.layers import RandomShift2DLayer, RandomScale2DLayer
 
 
 def default_model_creator():
     model = tf.keras.Sequential([
+        RandomScale2DLayer(minval=-10, maxval=10),
+        RandomShift2DLayer(minval=0, maxval=100, axis=1),
         tf.keras.layers.Flatten(),
         tf.keras.layers.Dense(units=320, activation='relu'),
         tf.keras.layers.Dropout(rate=0.05),
@@ -26,26 +26,3 @@ def default_model_creator():
     ])
 
     return model
-
-
-def conv_model_creator():
-    model = Sequential()
-    model.add(Conv1D(16, 3, padding='same', input_shape=(40611, 2)))
-    model.add(LeakyReLU())
-    model.add(Conv1D(16, 3, padding='same'))
-    model.add(LeakyReLU())
-    model.add(MaxPooling1D(pool_size=3))
-    model.add(Dropout(0.25))
-    model.add(Conv1D(16, 3, padding='same'))
-    model.add(LeakyReLU())
-    model.add(Conv1D(16, 3, padding='same'))
-    model.add(LeakyReLU())
-    model.add(MaxPooling1D(pool_size=3))
-    model.add(Dropout(0.25))
-    model.add(Flatten())
-    model.add(Dense(64))
-    model.add(LeakyReLU())
-    model.add(Dropout(0.5))
-    model.add(Dense(1, activation='sigmoid'))
-
-    return model

scripts/train2.py

@@ -4,7 +4,7 @@ from core.utils import load_ircam_hrirs_data, split_dataset, generate_signals, g
 import numpy as np
 import tensorflow as tf
 from core.logger import experiment_logger
-from core.model import default_model_creator, conv_model_creator
+from core.model import default_model_creator
 import pandas as pd
 from core.metrics import MeanAbsoluteAzimuthError
 
@@ -30,7 +30,7 @@ exp_config = ExperimentConfiguration({'n_augment': 1,
                                       'stimulus': 'dirac',  # possible values: ['dirac', 'any_sound']
                                       'train_subject_ids': ALL_SUBJECTS_EXCEPT_1059[0:30],
                                       'test_subject_ids': [1059],
-                                      'movement': 'deterministic',  # possible values: ['none', 'random', 'deterministic']
+                                      'movement': 'none',  # possible values: ['none', 'random', 'deterministic']
                                       'sessions': [0]})
 
 if __name__ == '__main__':
@@ -72,7 +72,8 @@ if __name__ == '__main__':
 
     valid_train_subject_ids = 'train_subject_ids' in exp_config.keys() and exp_config[
         'train_subject_ids'] is not None and len(exp_config['train_subject_ids']) > 0
-    valid_test_subject_ids = 'test_subject_ids' in exp_config.keys() and exp_config['test_subject_ids'] is not None and len(
+    valid_test_subject_ids = 'test_subject_ids' in exp_config.keys() and exp_config[
+        'test_subject_ids'] is not None and len(
         exp_config['test_subject_ids']) > 0
     if valid_train_subject_ids and valid_test_subject_ids:
         train_subject_ids = exp_config['train_subject_ids']
@@ -139,6 +140,11 @@ if __name__ == '__main__':
                                                                             sounds=test_sounds,
                                                                             n_samples=n_samples,
                                                                             n_augment=1)
+
+            # Expand dimensions of arrays to comply with the stanrdized shape BWHF (Batch, Width, Height, Features).
+            # Here, width=time axis, and Height=channel (left or right)
+            training_signals = np.expand_dims(training_signals, axis=-1)
+            test_signals = np.expand_dims(test_signals, axis=-1)
             model.fit(x=training_signals, y=training_labels, batch_size=batch_size, epochs=exp_config['n_epochs'],
                       validation_data=(test_signals, test_labels))