def create_dataset(self, mode):
# set directory for different modes and create respective datasets
if mode == 'train':
s_list = os.path.join(self.datapath, 'CM_protocol/cm_train.trn')
t = 15000
elif mode == 'dev':
s_list = os.path.join(self.datapath, 'CM_protocol/cm_develop.ndx')
t = 10000
elif mode == 'test':
s_list = os.path.join(self.datapath,
'CM_protocol/cm_evaluation.ndx')
t = 10000
else:
print('Invalid mode! specify one of train, dev, test.')
f = (self.sr, self.frame_size, self.frame_shift)
DLoad = Dataloader(file_dir=os.path.join(self.datapath, 'wav'),
speaker_list=s_list,
f_param=f,
thresh=t)
DLoad.create_dataset()
return DLoad
parser.add_argument('-samples', type=int, default=100, help="sampling")
parser.add_argument('-n_samples', type=int, default=50, help="train or test images")
parser.add_argument('-d_output', type=int, default=2, help="dimension of model ouput")
parser.add_argument('-batch_size', type=int, default=64, help="dimension of model ouput")
parser.add_argument('-n_filter', type=int, default=3, help="dimension of model ouput")
parser.add_argument('-model', type=str, default="point", help="training model")
parser.add_argument('-testpath', type=str, default="test.tfrecord", help="test path")
parser.add_argument('-tfrecordpath', type=str, default="train.tfrecord", help="dimension of model ouput")
parser.add_argument('-actv', type=str, default="relu", help="dimension of model ouput")
args = parser.parse_args()
dataloader = Dataloader(args)
if __name__ == "__main__":
dataset = dataloader.create_dataset()
optimizer = tf.train.AdamOptimizer(args.lr)
model_path = "{}_l_{}_b_{}_f_{}.h5".format(args.model, args.lr, args.batch_size, args.n_filter)
log_path = "{}".format(args.model)
if not os.path.exists(log_path):
os.mkdir(log_path)
callbacks = [ModelCheckpoint(filepath=model_path,
save_best_only=True,
monitor='loss',
verbose=2,
save_weights_only=True),
TensorBoard(log_dir=log_path)]