def batch_iterator(batch_size, k_samples, shuffle):
return BatchIterator(batch_size=batch_size // 4,
prepare=prepare_valid,
k_samples=k_samples,
shuffle=shuffle,
fill_last_batch=False)
def batch_iterator(batch_size, k_samples, shuffle):
return BatchIterator(batch_size=batch_size,
prepare=prepare_train,
k_samples=k_samples,
shuffle=shuffle)
if args.params:
dump_file = args.params
print("overwriting parameter dump file", os.path.basename(dump_file))
# compile network
net = model.build_model(batch_size=1)
# initialize neural network
my_net = Network(net, print_architecture=False)
# load model parameters network
my_net.load(dump_file)
# init batch iterator
bi = BatchIterator(batch_size=1,
k_samples=None,
shuffle=False,
prepare=model.prepare)
# iterate samples for prediction
print("Predicting on test set ...")
prediction_stats = []
y_true, y_probs, y_predicted = [], [], []
for i, (X, y) in enumerate(bi(data[args.set])):
print("Processing file %d / %d" % (i + 1, data[args.set].shape[0]),
end='\r')
sys.stdout.flush()
X_orig = X.copy()
# fix spectrogram lengths
if args.min_len: