def __init__(self):
super().__init__()
warnings.simplefilter(action='ignore', category=FutureWarning)
warnings.simplefilter(action='ignore', category=RuntimeWarning)
warnings.filterwarnings(action="ignore", category=DeprecationWarning)
# Deep Learning Models Configuration
K.set_image_dim_ordering('th')
print(K.image_data_format())
config = tf.ConfigProto(gpu_options=tf.GPUOptions(allow_growth=True))
session = tf.Session(config=config)
tensorflow_backend.set_session(session)
input_shape = (3, 10, 39)
self.cnn_model_path = ('../CNN/best_model_all_cnnBass.h5')
self.cnn_model = sequentialCNN(input_shape)
self.cnn_model.load_weights(self.cnn_model_path)
input_shape = (None, 98)
n_hidden = 60
self.rnn_model_path = ('../RNN/best_model_all_rnnBass.h5')
self.rnn_model = sequentialRNN(input_shape, 2, n_hidden)
self.rnn_model.load_weights(self.rnn_model_path)
self.batch_size = 512
listDB = [sys.argv[2]]
else:
listDB = ['Train', 'Validation']
for trainVal in listDB:
for mode in listItems:
print('Analizando mode {}'.format(mode))
midi_files = glob.glob('../MIDI/' + trainVal+'/'+mode+'/*.mid')
num_files = len(midi_files)
cnn_model_path = ('../CNN/best_model_' +
mode+'_cnnBass.h5')
cnn_model = sequentialCNN(input_shape)
cnn_model.load_weights(cnn_model_path)
input_shape = (None, 98)
rnn_model_path = ('../RNN/best_model_' + mode+'_rnnBass.h5')
rnn_model = sequentialRNN(input_shape,num_classes,n_hidden)
rnn_model.load_weights(rnn_model_path)
BM = BassManager()
#for midi_path in midi_files:
for i in range(num_files):
if i > len(midi_files)-1:
break
midi_path = midi_files[i]
tensorflow_backend.set_session(session)
# prepare callbacks
callbacks = defineCallBacks(best_model)
# load patches
if only_test == 1:
X_test, Y_test, classes = loadTestPatches(patches_file)
else:
X_train, Y_train, X_test, Y_test, classes = loadPatches(patches_file)
num_classes = len(classes)
input_shape = (1, X_test.shape[2], X_test.shape[3])
# CNN architecture defintion
model = sequentialCNN(input_shape, droprate, num_classes, p, d, f)
# training
if do_train == 1 and only_test == 0:
deleteWeights(best_model, last_model)
history = model.fit(X_train,
Y_train,
batch_size=batch_size,
epochs=epochs,
verbose=1,
validation_data=(X_test, Y_test),
shuffle=True,
callbacks=callbacks)
model.save_weights(last_model)
if plot_curves == 1:
plotCurves(history)