def testToNoteSequence(self):
converter = data.DrumsConverter(
pitch_classes=data.REDUCED_DRUM_PITCH_CLASSES,
slice_bars=None,
gap_bars=None,
steps_per_quarter=1,
roll_input=True,
roll_output=True,
max_tensors_per_notesequence=None)
_, output_tensors = converter.to_tensors(self.sequence)
sequences = converter.to_notesequences(output_tensors)
self.assertEqual(1, len(sequences))
expected_sequence = music_pb2.NoteSequence(ticks_per_quarter=220)
expected_sequence.tempos.add(qpm=120)
testing_lib.add_track_to_sequence(expected_sequence,
0, [(36, 80, 0, 0.5),
(42, 80, 0.5, 1.0),
(36, 80, 0.5, 1.0),
(38, 80, 2.0, 2.5),
(36, 80, 4.5, 5.0),
(38, 80, 6.5, 7.0),
(48, 80, 8.0, 8.5),
(49, 80, 8.0, 8.5),
(51, 80, 9.5, 10.0),
(38, 80, 16.5, 17.0),
(48, 80, 18.0, 18.5),
(49, 80, 18.0, 18.5),
(51, 80, 19.5, 20.0)],
is_drum=True)
for n in expected_sequence.notes:
n.instrument = 9
self.assertProtoEquals(expected_sequence, sequences[0])
def testIsTraining(self):
converter = data.DrumsConverter(
steps_per_quarter=1, slice_bars=1, max_tensors_per_notesequence=2)
self.is_training = True
self.assertEqual(2, len(converter.to_tensors(self.sequence)[0]))
converter.max_tensors_per_notesequence = None
self.assertEqual(5, len(converter.to_tensors(self.sequence)[0]))
def testIsTraining(self):
converter = data.DrumsConverter(
steps_per_quarter=1, slice_bars=1, max_tensors_per_notesequence=2)
converter.set_mode('train')
self.assertEqual(2, len(converter.to_tensors(self.sequence).inputs))
converter.max_tensors_per_notesequence = None
self.assertEqual(5, len(converter.to_tensors(self.sequence).inputs))
def testMaxOutputsPerNoteSequence(self):
converter = data.DrumsConverter(
steps_per_quarter=1, slice_bars=1, max_tensors_per_notesequence=2)
self.assertEqual(2, len(converter.to_tensors(self.sequence)[0]))
converter.max_tensors_per_notesequence = 3
self.assertEqual(3, len(converter.to_tensors(self.sequence)[0]))
converter.max_tensors_per_notesequence = 100
self.assertEqual(5, len(converter.to_tensors(self.sequence)[0]))
def testToNoteSequence(self):
converter = data.DrumsConverter(steps_per_quarter=1,
slice_bars=2,
max_tensors_per_notesequence=1)
tensors = converter.to_tensors(filter_instrument(self.sequence, 1))
sequences = converter.to_notesequences(tensors.outputs)
self.assertEqual(1, len(sequences))
expected_sequence = music_pb2.NoteSequence(ticks_per_quarter=220)
expected_sequence.tempos.add(qpm=120)
testing_lib.add_track_to_sequence(expected_sequence,
9, [(38, 80, 0.5, 1.0),
(48, 80, 2.0, 2.5),
(49, 80, 2.0, 2.5),
(51, 80, 3.5, 4.0)],
is_drum=True)
self.assertProtoEquals(expected_sequence, sequences[0])
def testSliced(self):
expected_sliced_events = [
([0], [0, 2], [], [],
[1], [], [], []),
([], [0], [], [],
[], [1], [], []),
([], [1], [], [],
[5, 7], [], [], [8]),
]
expected_silent_array = np.array([
[0, 0, 1, 1, 0, 1, 1, 1],
[1, 0, 1, 1, 1, 0, 1, 1],
[1, 0, 1, 1, 0, 1, 1, 0],
])
expected_output_tensors = np.zeros(
(len(expected_sliced_events), 8, len(data.REDUCED_DRUM_PITCH_CLASSES)),
np.bool)
for i, events in enumerate(expected_sliced_events):
for j, e in enumerate(events):
expected_output_tensors[i, j, e] = 1
converter = data.DrumsConverter(
pitch_classes=data.REDUCED_DRUM_PITCH_CLASSES,
slice_bars=2,
steps_per_quarter=1,
roll_input=True,
roll_output=True,
max_tensors_per_notesequence=None)
self.assertEqual(10, converter.input_depth)
self.assertEqual(9, converter.output_depth)
input_tensors, output_tensors, _ = converter.to_tensors(self.sequence)
self.assertArraySetsEqual(
np.append(
expected_output_tensors,
np.expand_dims(expected_silent_array, axis=2),
axis=2),
input_tensors)
self.assertArraySetsEqual(expected_output_tensors, output_tensors)
lstm_models.get_default_hparams(),
HParams(
batch_size=512,
max_seq_len=32, # 2 bars w/ 16 steps per bar
z_size=256,
enc_rnn_size=[512],
dec_rnn_size=[256, 256],
free_bits=48,
max_beta=0.2,
sampling_schedule='inverse_sigmoid',
sampling_rate=1000,
)),
note_sequence_augmenter=None,
data_converter=data.DrumsConverter(
max_bars=100, # Truncate long drum sequences before slicing.
slice_bars=2,
steps_per_quarter=4,
roll_input=True),
train_examples_path=None,
eval_examples_path=None,
)
CONFIG_MAP['cat-drums_2bar_big'] = Config(
model=MusicVAE(lstm_models.BidirectionalLstmEncoder(),
lstm_models.CategoricalLstmDecoder()),
hparams=merge_hparams(
lstm_models.get_default_hparams(),
HParams(
batch_size=512,
max_seq_len=32, # 2 bars w/ 16 steps per bar
z_size=512,
