# Get Train and test sets
train_set, test_set, chord_train_set, chord_test_set = data_class.get_ind_train_and_test_set(
train_set_size, test_set_size)
if chord_embed_method == 'embed':
chord_dim = chord_embedding_dim
elif chord_embed_method == 'onehot':
chord_dim = num_chords
elif chord_embed_method == 'int':
chord_dim = 1
if next_chord_feature:
chord_dim = chord_dim * 2
# Load model for chord embeddings
chord_embed_model = chord_model.Embed_Chord_Model(chord_model_path)
# Build Melody Model
print('creating model...')
model = Sequential()
# model.add(LSTM(lstm_size, batch_size=batch_size, input_shape=(step_size, new_num_notes+chord_dim+counter_size), stateful=True))
model.add(
LSTM(lstm_size,
batch_input_shape=(batch_size, step_size,
new_num_notes + chord_dim + counter_size),
stateful=True))
model.add(Dense(new_num_notes))
model.add(Activation('sigmoid'))
if optimizer == 'RMS': optimizer = RMSprop(lr=learning_rate)
if optimizer == 'Adam': optimizer = Adam(lr=learning_rate)
self.f_score.append(round(statistics.mean(f_scores), 2))
self.sequence_cross_entropy.append(
round(statistics.mean(sequence_cross_entropies), 2))
self.transition_cross_entropy.append(
round(statistics.mean(transition_cross_entropies), 2))
self.steady_state_cross_entropy.append(
round(statistics.mean(steady_state_cross_entropies), 2))
print(f'\nF-score:\t\t\t{self.f_score[-1]}')
print(f'Cross entropy:\t\t\t{self.sequence_cross_entropy[-1]}')
print(
f'Transition cross entropy:\t{self.transition_cross_entropy[-1]}'
)
print(
f'Steady state cross entropy:\t{self.steady_state_cross_entropy[-1]}'
)
if __name__ == '__main__':
if len(sys.argv) > 1:
if len(sys.argv) > 2:
evaluation = Evaluation(
interval=1,
chord_embed_model=chord_model.Embed_Chord_Model(sys.argv[1]))
evaluation.import_model(name=sys.argv[2])
evaluation.evaluate()
else:
print('Please specify the polyphonic model path.')
else:
print('Please specify the chord model path.')