def setup_classifier(load_weights_from):
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model_args = lib_rnn.set_default_args()
model = lib_rnn.create_RNN_model(model_args, load_weights_from)
if 0: # random test
label_index = model.predict(np.random.random((66, 12)))
print("Label index of a random feature: ", label_index)
exit("Complete test.")
return model
import torch
if 1: # my lib
import utils.lib_io as lib_io
import utils.lib_commons as lib_commons
import utils.lib_datasets as lib_datasets
import utils.lib_augment as lib_augment
import utils.lib_ml as lib_ml
import utils.lib_rnn as lib_rnn
# --------------------------------------------------
# --------------------------------------------------
# --------------------------------------------------
# Set arguments -------------------------
args = lib_rnn.set_default_args()
args.num_epochs = 15
args.learning_rate = 0.001
args.train_eval_test_ratio=[0.7, 0.3, 0.0]
args.do_data_augment = True
args.data_folder = "data/data_train/"
args.classes_txt = "config/classes.names"
args.load_weight_from = "weights/kaggle.ckpt"
args.finetune_model = True # If true, fix all parameters except the fc layer
args.save_model_to = 'checkpoints/' # Save model and log file
# Dataset --------------------------
# Get data's filenames and labels
file_paths, file_labels = lib_datasets.AudioDataset.load_classes_and_data_filenames(
def setup_classifier(load_weights_from):
model_args = lib_rnn.set_default_args()
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = lib_rnn.create_RNN_model(model_args, load_weights_from)
return model