def save_phrase_datasets(process_data_func=dataframe_to_note_set_progression,
prefix=""):
train_phrases, val_phrases, test_phrases = split_data_by_phrase(
"data/phrases.txt")
train_dataset = get_dataset(
"data/all_annotations.csv",
train_phrases,
process_data_func,
augment=True,
skip_short_phrases=SHORT_PHRASE_LEN,
skip_repetitions=SKIP_REPETITIONS,
skip_double_repetitions=SKIP_DOUBLE_REPETITIONS)
val_dataset = get_dataset("data/all_annotations.csv",
val_phrases,
process_data_func,
augment=False,
skip_short_phrases=SHORT_PHRASE_LEN,
skip_repetitions=SKIP_REPETITIONS,
skip_double_repetitions=SKIP_DOUBLE_REPETITIONS)
test_dataset = get_dataset("data/all_annotations.csv",
test_phrases,
process_data_func,
augment=False,
skip_short_phrases=SHORT_PHRASE_LEN,
skip_repetitions=SKIP_REPETITIONS,
skip_double_repetitions=SKIP_DOUBLE_REPETITIONS)
with open("data/train_{}phrases.txt".format(prefix), "w") as f:
f.write("\n".join(train_dataset))
with open("data/val_{}phrases.txt".format(prefix), "w") as f:
f.write("\n".join(val_dataset))
with open("data/test_{}phrases.txt".format(prefix), "w") as f:
f.write("\n".join(test_dataset))
def convert_phrases_to_root_progression():
phrase_list, _, _ = split_data_by_phrase("data/phrases.txt",
split_ratio=[1, 0, 0],
shuffle=False)
dataset = get_dataset("data/all_annotations.csv",
phrase_list,
dataframe_to_root_progression,
skip_short_phrases=SHORT_PHRASE_LEN,
skip_repetitions=SKIP_REPETITIONS,
skip_double_repetitions=SKIP_DOUBLE_REPETITIONS)
with open("data/phrases_root.txt", "w") as f:
f.write("\n".join(dataset))
def histogram_chords(chord_output, is_sorted=True):
if is_sorted:
process_data_func = dataframe_to_note_set_progression_sorted
else:
process_data_func = dataframe_to_note_set_progression
phrase_list, _, _ = split_data_by_phrase("data/phrases.txt",
split_ratio=[1, 0, 0],
shuffle=False)
dataset = get_dataset("data/all_annotations.csv",
phrase_list,
process_data_func,
skip_short_phrases=SHORT_PHRASE_LEN,
skip_repetitions=SKIP_REPETITIONS,
augment=True,
augment_func=transpose_phrase_to_c_maj_or_a_min)
unique_chords = set()
progression_list = []
for phrase in dataset:
progression = phrase.split(" ")
progression = convert_to_chord_name(progression,
output=chord_output,
is_sorted=is_sorted)
progression = progression.split(" ")
unique_chords.update(progression)
progression_list.append(progression)
unique_chords = list(unique_chords)
unique_chords = sorted(unique_chords)
n_rows = len(unique_chords)
n_cols = n_rows + 1
table = np.zeros((n_rows, n_cols))
for progression in progression_list:
for i in range(len(progression) - 1):
start_chord = progression[i]
end_chord = progression[i + 1]
start = unique_chords.index(start_chord)
end = unique_chords.index(end_chord)
table[start][end] += 1
last = unique_chords.index(progression[-1])
table[last][n_cols - 1] += 1
return table, unique_chords
def print_histogram_root(name="root_histogram"):
if TRANSPOSE:
augment_func = transpose_phrase_to_c_maj_or_a_min
else:
augment_func = None
phrase_list, _, _ = split_data_by_phrase("data/phrases.txt",
split_ratio=[1, 0, 0],
shuffle=False)
dataset = get_dataset("data/all_annotations.csv",
phrase_list,
dataframe_to_root_progression,
skip_short_phrases=SHORT_PHRASE_LEN,
skip_repetitions=SKIP_REPETITIONS,
augment=TRANSPOSE,
augment_func=augment_func)
unique_chords = set()
progression_list = []
for phrase in dataset:
progression = phrase.split(" ")
unique_chords.update(progression)
progression_list.append(progression)
unique_chords = list(unique_chords)
unique_chords = sorted(unique_chords)
n_rows = len(unique_chords)
counter = np.zeros((n_rows, ))
for progression in progression_list:
for root in progression:
index = unique_chords.index(root)
counter[index] += 1
plt.rcParams['figure.figsize'] = (20, 20)
plt.bar(np.arange(n_rows), counter)
print(n_rows)
plt.xticks(np.arange(n_rows), unique_chords, rotation=0, fontsize=10)
plt.savefig("figures/{}{} ({:d}).pdf".format(name, SUFFIX,
int(np.sum(counter))))
def convert_phrases():
phrase_list, _, _ = split_data_by_phrase("data/phrases.txt",
split_ratio=[1, 0, 0],
shuffle=False)
dataset = get_dataset("data/all_annotations.csv",
phrase_list,
dataframe_to_note_set_progression,
skip_short_phrases=SHORT_PHRASE_LEN,
skip_repetitions=SKIP_REPETITIONS,
skip_double_repetitions=SKIP_DOUBLE_REPETITIONS)
with open("data/phrases_note_set.txt", "w") as f:
f.write("\n".join(dataset))
progression_list = []
for phrase in dataset:
progression = phrase.split(" ")
progression = convert_to_chord_name(progression)
progression_list.append(progression)
with open("data/phrases_name.txt", "w") as f:
f.write("\n".join(progression_list))
def save_transposed_datasets():
train_phrases, val_phrases, test_phrases = split_data_by_phrase(
"data/phrases.txt")
process_data_func = dataframe_to_note_set_progression
augment_func = transpose_phrase_to_c_maj_or_a_min
train_dataset = get_dataset(
"data/all_annotations.csv",
train_phrases,
process_data_func,
augment=True,
skip_short_phrases=SHORT_PHRASE_LEN,
skip_repetitions=SKIP_REPETITIONS,
skip_double_repetitions=SKIP_DOUBLE_REPETITIONS,
augment_func=augment_func)
val_dataset = get_dataset("data/all_annotations.csv",
val_phrases,
process_data_func,
augment=True,
skip_short_phrases=SHORT_PHRASE_LEN,
skip_repetitions=SKIP_REPETITIONS,
skip_double_repetitions=SKIP_DOUBLE_REPETITIONS,
augment_func=augment_func)
test_dataset = get_dataset("data/all_annotations.csv",
test_phrases,
process_data_func,
augment=True,
skip_short_phrases=SHORT_PHRASE_LEN,
skip_repetitions=SKIP_REPETITIONS,
skip_double_repetitions=SKIP_DOUBLE_REPETITIONS,
augment_func=augment_func)
with open("data/train_transposed_phrases.txt", "w") as f:
f.write("\n".join(train_dataset))
with open("data/val_transposed_phrases.txt", "w") as f:
f.write("\n".join(val_dataset))
with open("data/test_transposed_phrases.txt", "w") as f:
f.write("\n".join(test_dataset))
def histogram_transposed_local_key_root(
process_data_func=dataframe_to_root_progression):
phrase_list, _, _ = split_data_by_phrase("data/phrases.txt",
split_ratio=[1, 0, 0],
shuffle=False)
dataset, modulation_list = get_dataset_and_modulation(
"data/all_annotations.csv",
phrase_list,
process_data_func,
skip_short_phrases=SHORT_PHRASE_LEN,
skip_repetitions=SKIP_REPETITIONS)
unique_chords = set()
progression_list = []
for phrase in dataset:
progression = phrase.split(" ")
unique_chords.update(progression)
progression_list.append(progression)
unique_chords = list(unique_chords)
unique_chords = sorted(unique_chords)
n_rows = len(unique_chords)
n_cols = n_rows + 1
table = np.zeros((n_rows, n_cols))
for progression in progression_list:
for i in range(len(progression) - 1):
start_chord = progression[i]
end_chord = progression[i + 1]
start = unique_chords.index(start_chord)
end = unique_chords.index(end_chord)
table[start][end] += 1
last = unique_chords.index(progression[-1])
table[last][n_cols - 1] += 1
return table, unique_chords, modulation_list