gen_log_dir = 'logs/mt_decoder/generate_' + current_time + '/generate'
gen_summary_writer = tf.summary.create_file_writer(gen_log_dir)
if mode == 'enc-dec':
print(">> generate with original seq2seq wise... beam size is {}".format(
beam))
mt = MusicTransformer(embedding_dim=256,
vocab_size=par.vocab_size,
num_layer=6,
max_seq=2048,
dropout=0.2,
debug=False,
loader_path=load_path)
else:
print(">> generate with decoder wise... beam size is {}".format(beam))
mt = MusicTransformerDecoder(loader_path=load_path)
inputs = encode_midi('dataset/midi/BENABD10.mid')
with gen_summary_writer.as_default():
result = mt.generate(inputs[:10], beam=beam, length=length, tf_board=True)
for i in result:
print(i)
if mode == 'enc-dec':
decode_midi(list(inputs[-1 * par.max_seq:]) + list(result[1:]),
file_path=save_path)
else:
decode_midi(result, file_path=save_path)
loader_path=load_path)
mt.compile(optimizer=opt, loss=callback.transformer_dist_train_loss)
# Train Start
for e in range(epochs):
mt.reset_metrics()
for b in range(len(dataset.files) // batch_size):
try:
batch_x, batch_y = dataset.seq2seq_batch(batch_size, max_seq)
except:
continue
result_metrics = mt.train_on_batch(batch_x, batch_y)
if b % 100 == 0:
eval_x, eval_y = dataset.seq2seq_batch(batch_size, max_seq, 'eval')
print('eval_x', len(eval_x[0]), eval_x)
print('eval_y', len(eval_y[0]), eval_y)
# print('generating ...',len(eval_x[0]))
gen_res = mt.generate(eval_x[0][:1024], beam=3, length=1024)
print('generated sequence: ', gen_res)
midi0 = decode_midi(
gen_res[0],
file_path='result/midi/result-{}-{}.mid'.format(e, b))
eval_result_metrics = mt.evaluate(eval_x, eval_y)
mt.save(save_path, e)
print('\n====================================================')
print('Epoch/Batch: {}/{}'.format(e, b))
print('Train >>>> Loss: {:6.6}, Accuracy: {}'.format(
result_metrics[0], result_metrics[1]))
print('Eval >>>> Loss: {:6.6}, Accuracy: {}'.format(
eval_result_metrics[0], eval_result_metrics[1]))
