random.seed(c.train_config["seed"])
torch.manual_seed(c.train_config["seed"])
torch.cuda.manual_seed(c.train_config["seed"])
np.random.seed(c.train_config["seed"])
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False

self.c = c
self.ap = ap
self.train = train
self.test = test
self.test_insert_noise = test_insert_noise
self.num_test_additive_noise = num_test_additive_noise
self.num_test_specaug = num_test_specaug
self.dataset_csv = \
  c.dataset["train_csv"] if train else c.dataset["eval_csv"]

assert os.path.isfile(self.dataset_csv), \
  "Test or Train CSV file don't exists! Fix it in config.json")

accepted_tc = [ 'overlapping', 'padding', 'one_window' ]
assert self.c.dataset['temporal_control'] in accepted_tc), \
  "You cannot use the padding_with_max_length option with the \
   split_wav_using_overlapping option, disable one of them !!")

self.control_class = c.dataset['control_class']
self.patient_class = c.dataset['patient_class']

self.dataset_list = \
  pd.read_csv(self.dataset_csv, sep=',') \
  .replace({'?': -1}) \
  .replace({'negative': self.control_class}, regex=True) \
  .replace({'positive': self.patient_class}, regex=True) \
  .values

# Setup

config_path = ValidPath.from_str("/app/spira/spira.json")
config = load_config(config_path)

operation_mode = OperationMode.TRAIN
randomizer = initialize_random(config, operation_mode)

# Data Loading

patients_paths = read_valid_paths_from_csv(config.patients_csv)
controls_paths = read_valid_paths_from_csv(config.controls_csv)
noises_paths = read_valid_paths_from_csv(config.noises_csv)

patients_inputs = Audios.load(
  patients_paths, config.audio, config.dataset
)
controls_inputs = Audios.load(
  controls_paths, config.audio, config.dataset
)
noises = Audios.load(noises_paths, config)
  noises_path, config.audio, config.dataset
)

# Feature Engineering

audio_processor = create_audio_processor(config.audio)

patient_feature_transformer = create_audio_feature_transformer(
    randomizer, audio_processor, config, noises,
)
control_feature_transformer = create_audio_feature_transformer(
    randomizer, audio_processor, config, noises,
)