def slice_cqt(row, window_length):
"""Generate slices of CQT observations.
Parameters
----------
row : pd.Series
Row from a features dataframe.
window_length : int
Length of the CQT slice in time.
Yields
------
x_obs : np.ndarray
Slice of cqt data.
meta : dict
Metadata corresponding to the observation.
"""
try:
data = np.load(row.features)['cqt']
except IOError as derp:
print("Failed reading row: {}\n\n{}".format(row.to_dict(), derp))
raise derp
num_obs = data.shape[1]
# Break the remainder out into a subfunction for reuse with embedding
# sampling.
idx = np.random.permutation(num_obs) if num_obs > 0 else None
if idx is None:
raise ValueError("Misshapen CQT ({}) - {}".format(
data.shape, row.to_dict()))
np.random.shuffle(idx)
counter = 0
meta = dict(instrument=row.instrument,
note_number=row.note_number,
fcode=row.fcode)
while num_obs > 0:
n = idx[counter]
obs = utils.padded_slice_ndarray(data, n, length=window_length, axis=1)
obs = obs[np.newaxis, ...]
meta['idx'] = n
yield obs, meta
counter += 1
if counter >= len(idx):
np.random.shuffle(idx)
counter = 0
def slice_cqt(row, window_length):
"""Generate slices of CQT observations.
Parameters
----------
row : pd.Series
Row from a features dataframe.
window_length : int
Length of the CQT slice in time.
Yields
------
x_obs : np.ndarray
Slice of cqt data.
meta : dict
Metadata corresponding to the observation.
"""
try:
data = np.load(row.features)['cqt']
except IOError as derp:
print("Failed reading row: {}\n\n{}".format(row.to_dict(), derp))
raise derp
num_obs = data.shape[1]
# Break the remainder out into a subfunction for reuse with embedding
# sampling.
idx = np.random.permutation(num_obs) if num_obs > 0 else None
if idx is None:
raise ValueError(
"Misshapen CQT ({}) - {}".format(data.shape, row.to_dict()))
np.random.shuffle(idx)
counter = 0
meta = dict(instrument=row.instrument, note_number=row.note_number,
fcode=row.fcode)
while num_obs > 0:
n = idx[counter]
obs = utils.padded_slice_ndarray(data, n, length=window_length, axis=1)
obs = obs[np.newaxis, ...]
meta['idx'] = n
yield obs, meta
counter += 1
if counter >= len(idx):
np.random.shuffle(idx)
counter = 0
def slice_cqt_weighted(row, window_length):
"""Generate slices of CQT observations.
Parameters
----------
row : pd.Series
Row from a features dataframe.
window_length : int
Length of the CQT slice in time.
Yields
------
x_obs : np.ndarray
Slice of cqt data.
meta : dict
Metadata corresponding to the observation.
"""
try:
data = np.load(row.features)['cqt']
except IOError as derp:
print("Failed reading row: {}\n\n{}".format(row.to_dict(), derp))
raise derp
# Create an index likelihood as a function of amplitude
weights = data.squeeze().sum(axis=-1)
weights /= weights.sum()
meta = dict(instrument=row.instrument,
note_number=row.note_number,
fcode=row.fcode)
while True:
n = np.random.multinomial(1, weights).argmax()
obs = utils.padded_slice_ndarray(data, n, length=window_length, axis=1)
obs = obs[np.newaxis, ...]
meta['idx'] = n
yield obs, meta
def slice_cqt_weighted(row, window_length):
"""Generate slices of CQT observations.
Parameters
----------
row : pd.Series
Row from a features dataframe.
window_length : int
Length of the CQT slice in time.
Yields
------
x_obs : np.ndarray
Slice of cqt data.
meta : dict
Metadata corresponding to the observation.
"""
try:
data = np.load(row.features)['cqt']
except IOError as derp:
print("Failed reading row: {}\n\n{}".format(row.to_dict(), derp))
raise derp
# Create an index likelihood as a function of amplitude
weights = data.squeeze().sum(axis=-1)
weights /= weights.sum()
meta = dict(instrument=row.instrument, note_number=row.note_number,
fcode=row.fcode)
while True:
n = np.random.multinomial(1, weights).argmax()
obs = utils.padded_slice_ndarray(data, n, length=window_length, axis=1)
obs = obs[np.newaxis, ...]
meta['idx'] = n
yield obs, meta
def test_padded_slice_ndarray():
x_in = np.arange(64).reshape(2, 4, 8)
np.testing.assert_equal(
utils.padded_slice_ndarray(x_in, idx=0, length=4, axis=0),
np.concatenate([np.zeros([2, 4, 8]), x_in[:2, ...]], axis=0))