def test_find_with_extension():
root = tempfile.mkdtemp()
files = [[root, 'file1.txt'],
[root, 'sub1', 'file2.txt'],
[root, 'sub1', 'sub2', 'file3.txt'],
[root, 'sub1', 'sub2', 'sub3', 'file4.txt']]
files = [os.sep.join(_) for _ in files]
badfiles = [_.replace('.txt', '.csv') for _ in files]
# Create all the necessary directories
util.smkdirs(os.path.dirname(files[-1]))
# Create the dummy files
for fname in files + badfiles:
with open(fname, 'w') as _:
pass
def __test(level, sort):
results = util.find_with_extension(root, 'txt', depth=level, sort=sort)
eq_(sorted(results), sorted(files[:level]))
for level in [1, 2, 3, 4]:
for sort in [False, True]:
yield __test, level, sort
# Cleanup
for fname, badfname in zip(files[::-1], badfiles[::-1]):
os.remove(fname)
os.remove(badfname)
os.rmdir(os.path.dirname(fname))
def root_and_files():
root = tempfile.mkdtemp()
files = [[root, 'file1.txt'], [root, 'sub1', 'file2.txt'],
[root, 'sub1', 'sub2', 'file3.txt'],
[root, 'sub1', 'sub2', 'sub3', 'file4.txt']]
files = [os.sep.join(_) for _ in files]
badfiles = [_.replace('.txt', '.csv') for _ in files]
# Create all the necessary directories
util.smkdirs(os.path.dirname(files[-1]))
# Create the dummy files
for fname in files + badfiles:
with open(fname, 'w'):
pass
yield root, files
for fname, badfname in zip(files[::-1], badfiles[::-1]):
os.remove(fname)
os.remove(badfname)
os.rmdir(os.path.dirname(fname))
def root_and_files():
root = tempfile.mkdtemp()
files = [[root, 'file1.txt'],
[root, 'sub1', 'file2.txt'],
[root, 'sub1', 'sub2', 'file3.txt'],
[root, 'sub1', 'sub2', 'sub3', 'file4.txt']]
files = [os.sep.join(_) for _ in files]
badfiles = [_.replace('.txt', '.csv') for _ in files]
# Create all the necessary directories
util.smkdirs(os.path.dirname(files[-1]))
# Create the dummy files
for fname in files + badfiles:
with open(fname, 'w'):
pass
yield root, files
for fname, badfname in zip(files[::-1], badfiles[::-1]):
os.remove(fname)
os.remove(badfname)
os.rmdir(os.path.dirname(fname))
def make_muda(presets):
'''Construct a MUDA pitch shifter'''
drc = muda.deformers.DynamicRangeCompression(preset=presets)
smkdirs(OUTPUT_PATH)
with open(os.path.join(OUTPUT_PATH, 'muda_drc.pkl'), 'wb') as fd:
pickle.dump(drc, fd)
return drc
def make_muda(stretch, n_stretch):
'''Construct a MUDA time stretcher'''
shifter = muda.deformers.LogspaceTimeStretch(n_samples=n_stretch, lower=-stretch, upper=stretch)
smkdirs(OUTPUT_PATH)
with open(os.path.join(OUTPUT_PATH, 'muda.pkl'), 'wb') as fd:
pickle.dump(shifter, fd)
return shifter
def make_muda(stretch, n_stretch):
'''Construct a MUDA time stretcher'''
shifter = muda.deformers.LogspaceTimeStretch(n_samples=n_stretch,
lower=-stretch,
upper=stretch)
smkdirs(OUTPUT_PATH)
with open(os.path.join(OUTPUT_PATH, 'muda.pkl'), 'wb') as fd:
pickle.dump(shifter, fd)
return shifter
def make_muda(n_semitones):
'''Construct a MUDA pitch shifter'''
tones = []
for n in range(1, n_semitones + 1):
tones.extend([-n, n])
shifter = muda.deformers.PitchShift(n_semitones=tones)
smkdirs(OUTPUT_PATH)
with open(os.path.join(OUTPUT_PATH, 'muda.pkl'), 'wb') as fd:
pickle.dump(shifter, fd)
return shifter
def test_smkdirs():
root = tempfile.mkdtemp()
my_dirs = [root, 'level1', 'level2', 'level3']
try:
target = os.sep.join(my_dirs)
util.smkdirs(target)
for i in range(1, len(my_dirs)):
tmpdir = os.sep.join(my_dirs[:i])
assert os.path.exists(tmpdir)
assert os.path.isdir(tmpdir)
finally:
for i in range(len(my_dirs), 0, -1):
tmpdir = os.sep.join(my_dirs[:i])
os.rmdir(tmpdir)
def test_smkdirs():
root = tempfile.mkdtemp()
my_dirs = [root, 'level1', 'level2', 'level3']
try:
target = os.sep.join(my_dirs)
util.smkdirs(target)
for i in range(1, len(my_dirs)):
tmpdir = os.sep.join(my_dirs[:i])
assert os.path.exists(tmpdir)
assert os.path.isdir(tmpdir)
finally:
for i in range(len(my_dirs), 0, -1):
tmpdir = os.sep.join(my_dirs[:i])
os.rmdir(tmpdir)
use_tqdm=True,
weak_from_strong=True)
# Save results to disk
results_file = os.path.join(OUTPUT_PATH, modelid, '{:02d}'.format(fold),
'results.json')
with open(results_file, 'w') as fp:
json.dump(results, fp, indent=2)
print('Done!')
if __name__ == '__main__':
params = process_arguments(sys.argv[1:])
smkdirs(OUTPUT_PATH)
# Get current directory
cwd = os.getcwd()
# Get directory where git repo lives
curfilePath = os.path.relpath(milsed.__file__)
curDir = os.path.abspath(os.path.join(curfilePath, os.pardir))
parentDir = os.path.abspath(os.path.join(curDir, os.pardir))
# Change to the repo directory
os.chdir(parentDir)
# Get GIT version
version = milsed.utils.increment_version(
os.path.join(OUTPUT_PATH, 'version.txt'))
# Return to working dir
os.chdir(cwd)
duration,
modelid,
use_orig_duration=True)
# Save results to disk
results_file = os.path.join(OUTPUT_PATH, modelid, 'results.json')
with open(results_file, 'w') as fp:
json.dump(results, fp, indent=2)
print('Done!')
if __name__ == '__main__':
params = process_arguments(sys.argv[1:])
smkdirs(OUTPUT_PATH)
# Get current directory
cwd = os.getcwd()
# Get directory where git repo lives
curfilePath = os.path.relpath(milsed.__file__)
curDir = os.path.abspath(os.path.join(curfilePath, os.pardir))
parentDir = os.path.abspath(os.path.join(curDir, os.pardir))
# Change to the repo directory
os.chdir(parentDir)
# Get GIT version
version = milsed.utils.increment_version(
os.path.join(OUTPUT_PATH, 'version.txt'))
# Return to working dir
os.chdir(cwd)
with open(os.path.join(OUTPUT_PATH, 'pump.pkl'), 'wb') as fd:
pickle.dump(pump, fd)
return pump
def convert(aud, jam, pump, outdir):
data = pump.transform(aud, jam)
fname = os.path.extsep.join(
[os.path.join(outdir, crema.utils.base(aud)), 'h5'])
crema.utils.save_h5(fname, **data)
if __name__ == '__main__':
params = process_arguments(sys.argv[1:])
smkdirs(OUTPUT_PATH)
smkdirs(params.output_path)
print('{}: pre-processing'.format(__doc__))
print(params)
pump = make_pump(params.sr, params.hop_length, params.n_octaves)
stream = tqdm(crema.utils.get_ann_audio(params.input_path),
desc='Converting training data')
Parallel(n_jobs=params.n_jobs)(
delayed(convert)(aud, ann, pump, params.output_path)
for aud, ann in stream)
if params.augment_path:
stream = tqdm(crema.utils.get_ann_audio(params.augment_path),
desc='Converting augmented data')
cb.append(K.callbacks.EarlyStopping(patience=early_stopping,
verbose=1,
monitor=monitor))
# Fit the model
model.fit_generator(gen_train, epoch_size, epochs,
validation_data=gen_val,
validation_steps=validation_size,
callbacks=cb)
if __name__ == '__main__':
params = process_arguments(sys.argv[1:])
smkdirs(OUTPUT_PATH)
version = crema.utils.increment_version(os.path.join(OUTPUT_PATH,
'version.txt'))
print('{}: training'.format(__doc__))
print('Model version: {}'.format(version))
print(params)
train(params.working,
params.max_samples, params.duration,
params.rate,
params.batch_size,
params.epochs, params.epoch_size,
params.validation_size,
params.early_stopping,
with open(os.path.join(OUTPUT_PATH, 'pump.pkl'), 'wb') as fd:
pickle.dump(pump, fd)
return pump
def convert(aud, jam, pump, outdir):
data = pump.transform(aud, jam)
fname = os.path.extsep.join([os.path.join(outdir, crema.utils.base(aud)),
'h5'])
crema.utils.save_h5(fname, **data)
if __name__ == '__main__':
params = process_arguments(sys.argv[1:])
smkdirs(OUTPUT_PATH)
smkdirs(params.output_path)
print('{}: pre-processing'.format(__doc__))
print(params)
pump = make_pump(params.sr, params.hop_length, params.n_octaves)
stream = tqdm(crema.utils.get_ann_audio(params.input_path),
desc='Converting training data')
Parallel(n_jobs=params.n_jobs)(delayed(convert)(aud, ann,
pump,
params.output_path)
for aud, ann in stream)
if params.augment_path:
stream = tqdm(crema.utils.get_ann_audio(params.augment_path),
cb.append(
K.callbacks.EarlyStopping(patience=early_stopping,
verbose=1,
monitor=monitor))
# Fit the model
model.fit_generator(gen_train,
epoch_size,
epochs,
validation_data=gen_val,
validation_steps=validation_size,
callbacks=cb)
if __name__ == '__main__':
params = process_arguments(sys.argv[1:])
smkdirs(OUTPUT_PATH)
version = crema.utils.increment_version(
os.path.join(OUTPUT_PATH, 'version.txt'))
print('{}: training'.format(__doc__))
print('Model version: {}'.format(version))
print(params)
train(params.working, params.max_samples, params.duration, params.rate,
params.batch_size, params.epochs, params.epoch_size,
params.early_stopping, params.reduce_lr, params.seed)