def compose_and_evaluate_piece(rl_tuner,
stat_dict,
composition_length=32,
key=None,
tonic_note=rl_tuner_ops.C_MAJOR_TONIC,
sample_next_obs=True):
"""Composes a piece using the model, stores statistics about it in a dict.
Args:
rl_tuner: An RLTuner object.
stat_dict: A dictionary storing statistics about a series of compositions.
composition_length: The number of beats in the composition.
key: The numeric values of notes belonging to this key. Defaults to
C-major if not provided.
tonic_note: The tonic/1st note of the desired key.
sample_next_obs: If True, each note will be sampled from the model's
output distribution. If False, each note will be the one with maximum
value according to the model.
Returns:
A dictionary updated to include statistics about the composition just
created.
"""
last_observation = rl_tuner.prime_internal_models()
rl_tuner.reset_composition()
for _ in range(composition_length):
if sample_next_obs:
action, new_observation, _ = rl_tuner.action(
last_observation,
0,
enable_random=False,
sample_next_obs=sample_next_obs)
else:
action, _ = rl_tuner.action(
last_observation,
0,
enable_random=False,
sample_next_obs=sample_next_obs)
new_observation = action
obs_note = np.argmax(new_observation)
# Compute note by note stats as it composes.
stat_dict = add_interval_stat(rl_tuner, new_observation, stat_dict, key=key)
stat_dict = add_in_key_stat(obs_note, stat_dict, key=key)
stat_dict = add_tonic_start_stat(
rl_tuner, obs_note, stat_dict, tonic_note=tonic_note)
stat_dict = add_repeating_note_stat(rl_tuner, obs_note, stat_dict)
stat_dict = add_motif_stat(rl_tuner, new_observation, stat_dict)
stat_dict = add_repeated_motif_stat(rl_tuner, new_observation, stat_dict)
stat_dict = add_leap_stats(rl_tuner, new_observation, stat_dict)
rl_tuner.composition.append(np.argmax(new_observation))
rl_tuner.beat += 1
last_observation = new_observation
for lag in [1, 2, 3]:
stat_dict['autocorrelation' + str(lag)].append(
rl_tuner_ops.autocorrelate(rl_tuner.composition, lag))
add_high_low_unique_stats(rl_tuner, stat_dict)
return stat_dict
def compose_and_evaluate_piece(rl_tuner,
stat_dict,
composition_length=32,
key=None,
tonic_note=rl_tuner_ops.C_MAJOR_TONIC,
sample_next_obs=True):
"""Composes a piece using the model, stores statistics about it in a dict.
Args:
rl_tuner: An RLTuner object.
stat_dict: A dictionary storing statistics about a series of compositions.
composition_length: The number of beats in the composition.
key: The numeric values of notes belonging to this key. Defaults to
C-major if not provided.
tonic_note: The tonic/1st note of the desired key.
sample_next_obs: If True, each note will be sampled from the model's
output distribution. If False, each note will be the one with maximum
value according to the model.
Returns:
A dictionary updated to include statistics about the composition just
created.
"""
last_observation = rl_tuner.prime_internal_models()
rl_tuner.reset_composition()
for _ in range(composition_length):
if sample_next_obs:
action, new_observation, _ = rl_tuner.action(
last_observation,
0,
enable_random=False,
sample_next_obs=sample_next_obs)
else:
action, _ = rl_tuner.action(
last_observation,
0,
enable_random=False,
sample_next_obs=sample_next_obs)
new_observation = action
obs_note = np.argmax(new_observation)
# Compute note by note stats as it composes.
stat_dict = add_interval_stat(rl_tuner, new_observation, stat_dict, key=key)
stat_dict = add_in_key_stat(obs_note, stat_dict, key=key)
stat_dict = add_tonic_start_stat(
rl_tuner, obs_note, stat_dict, tonic_note=tonic_note)
stat_dict = add_repeating_note_stat(rl_tuner, obs_note, stat_dict)
stat_dict = add_motif_stat(rl_tuner, new_observation, stat_dict)
stat_dict = add_repeated_motif_stat(rl_tuner, new_observation, stat_dict)
stat_dict = add_leap_stats(rl_tuner, new_observation, stat_dict)
rl_tuner.composition.append(np.argmax(new_observation))
rl_tuner.beat += 1
last_observation = new_observation
for lag in [1, 2, 3]:
stat_dict['autocorrelation' + str(lag)].append(
rl_tuner_ops.autocorrelate(rl_tuner.composition, lag))
add_high_low_unique_stats(rl_tuner, stat_dict)
return stat_dict