def emotions_train_tf(classifier: Type[AbstractClassifier],
model_path: str,
epochs: Optional[int] = None):
filtered = []
tracks = get_tracks(TRACKS_TYPE_SINGLE_EMOTION_ONLY)
for _, track in tracks.iterrows():
if os.path.isfile(
get_single_track_features_path(
track[Column.YM_TRACK_ID.value])):
filtered.append(track)
tracks = pd.DataFrame(filtered, columns=tracks.columns)
tracks[Column.EMOTIONS.value] = tracks[Column.EMOTIONS.value].apply(
lambda x: x.split('|'))
track_ids = tracks[Column.YM_TRACK_ID.value].tolist()
labels = tracks.set_index([Column.YM_TRACK_ID.value
]).to_dict('dict')[Column.EMOTIONS.value]
model = classifier.build_model(N_FEATURES)
classifier.train_model(model=model,
track_ids=track_ids,
labels=labels,
output_path=model_path,
epochs=epochs,
tracks_generator_class=TracksGenerator)
def _on_init_end(self):
self.track_ids_with_minute_indexes = []
for track_id in self.track_ids:
for minute_index in range(
len(np.load(get_single_track_features_path(track_id)))):
self.track_ids_with_minute_indexes.append(
(track_id, minute_index))
def __call__(self, sample: tuple[int, Any]) -> tuple[Tensor, Any]:
track_id, label = sample
if dump_file_exists(track_id, self.dump_dir):
return torch.tensor(np.load(get_single_track_features_path(track_id))), label
features = extract_features(track_id)
save_single_track_features(track_id, features, self.dump_dir)
if len(features) == 0:
return torch.zeros((self.chunk_size, self.n_mels)), label
return torch.tensor(features), label
def get_track_ids_and_labels():
filtered = []
tracks = get_tracks(TRACKS_TYPE_SINGLE_EMOTION_ONLY)
for _, track in tracks.iterrows():
if os.path.isfile(
get_single_track_features_path(
track[Column.YM_TRACK_ID.value])):
filtered.append(track)
tracks = pd.DataFrame(filtered, columns=tracks.columns)
tracks[Column.EMOTIONS.value] = tracks[Column.EMOTIONS.value].apply(
lambda x: x.split('|'))
track_ids = tracks[Column.YM_TRACK_ID.value].tolist()
labels = tracks.set_index([Column.YM_TRACK_ID.value
]).to_dict('dict')[Column.EMOTIONS.value]
return track_ids, labels
def _data_generation(
self, batch_items: list[int]) -> tuple[np.ndarray, np.ndarray]:
"""
X.shape: (n_samples, *dim)
y.shape: (n_samples, len(label_names))
:param batch_items:
:return: X, y
"""
X_list = []
y_labels = []
for track_id in batch_items:
features = np.load(get_single_track_features_path(track_id))
X_list.append(features[0])
y_labels.append(self.labels[track_id])
return np.array(X_list), self.mlb.transform(y_labels)