def test_gpu_stump_identical_subsequence_A_B_join():
identical = np.random.rand(8)
T_A = np.random.rand(20)
T_B = np.random.rand(20)
T_A[1:1 + identical.shape[0]] = identical
T_B[11:11 + identical.shape[0]] = identical
m = 3
ref_mp = naive.stamp(T_B, m, T_B=T_A)
comp_mp = gpu_stump(T_B, m, T_A, ignore_trivial=False)
naive.replace_inf(ref_mp)
naive.replace_inf(comp_mp)
npt.assert_almost_equal(
ref_mp[:, 0], comp_mp[:, 0],
decimal=config.STUMPY_TEST_PRECISION) # ignore indices
# comp_mp = gpu_stump(pd.Series(T_B), m, pd.Series(T_A), ignore_trivial=False)
# naive.replace_inf(comp_mp)
# npt.assert_almost_equal(
# ref_mp[:, 0], comp_mp[:, 0], decimal=config.STUMPY_TEST_PRECISION
# ) # ignore indices
# Swap inputs
ref_mp = naive.stamp(T_A, m, T_B=T_B)
comp_mp = gpu_stump(T_A, m, T_B, ignore_trivial=False)
naive.replace_inf(ref_mp)
naive.replace_inf(comp_mp)
npt.assert_almost_equal(
ref_mp[:, 0], comp_mp[:, 0],
decimal=config.STUMPY_TEST_PRECISION) # ignore indices
def test_two_constant_subsequences_A_B_join():
T_A = np.array([0, 0, 0, 0, 0, 1], dtype=np.float64)
T_B = np.concatenate(
(np.zeros(20, dtype=np.float64), np.ones(5, dtype=np.float64)))
m = 3
left = np.array(
[utils.naive_mass(Q, T_A, m) for Q in core.rolling_window(T_B, m)],
dtype=object)
right = gpu_stump(T_A, m, T_B, ignore_trivial=False)
utils.replace_inf(left)
utils.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0]) # ignore indices
right = gpu_stump(pd.Series(T_A), m, pd.Series(T_B), ignore_trivial=False)
utils.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0]) # ignore indices
# Swap inputs
left = np.array(
[utils.naive_mass(Q, T_B, m) for Q in core.rolling_window(T_A, m)],
dtype=object)
right = gpu_stump(T_B, m, T_A, ignore_trivial=False)
utils.replace_inf(left)
utils.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0]) # ignore indices
right = gpu_stump(pd.Series(T_B), m, pd.Series(T_A), ignore_trivial=False)
utils.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0]) # ignore indices
def test_gpu_stump_self_join_larger_window(T_A, T_B):
for m in [8, 16, 32]:
if len(T_B) > m:
zone = int(np.ceil(m / 4))
left = np.array(
[
utils.naive_mass(Q, T_B, m, i, zone, True)
for i, Q in enumerate(core.rolling_window(T_B, m))
],
dtype=object,
)
right = gpu_stump(T_B,
m,
ignore_trivial=True,
threads_per_block=THREADS_PER_BLOCK)
utils.replace_inf(left)
utils.replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(
pd.Series(T_B),
m,
ignore_trivial=True,
threads_per_block=THREADS_PER_BLOCK,
)
utils.replace_inf(right)
npt.assert_almost_equal(left, right)
def test_parallel_gpu_stump_self_join(T_A, T_B):
device_ids = [device.id for device in cuda.list_devices()]
if len(T_B) > 10:
m = 3
zone = int(np.ceil(m / 4))
left = np.array(
[
utils.naive_mass(Q, T_B, m, i, zone, True)
for i, Q in enumerate(core.rolling_window(T_B, m))
],
dtype=object,
)
right = gpu_stump(
T_B,
m,
ignore_trivial=True,
threads_per_block=THREADS_PER_BLOCK,
device_id=device_ids,
)
utils.replace_inf(left)
utils.replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(
pd.Series(T_B),
m,
ignore_trivial=True,
threads_per_block=THREADS_PER_BLOCK,
device_id=device_ids,
)
utils.replace_inf(right)
npt.assert_almost_equal(left, right)
def test_parallel_gpu_stump_A_B_join(T_A, T_B):
device_ids = [device.id for device in cuda.list_devices()]
if len(T_B) > 10:
m = 3
left = np.array(
[utils.naive_mass(Q, T_A, m) for Q in core.rolling_window(T_B, m)],
dtype=object,
)
right = gpu_stump(
T_A,
m,
T_B,
ignore_trivial=False,
threads_per_block=THREADS_PER_BLOCK,
device_id=device_ids,
)
utils.replace_inf(left)
utils.replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(
pd.Series(T_A),
m,
pd.Series(T_B),
ignore_trivial=False,
threads_per_block=THREADS_PER_BLOCK,
device_id=device_ids,
)
utils.replace_inf(right)
npt.assert_almost_equal(left, right)
def test_gpu_stump_nan_inf_A_B_join(
T_A, T_B, substitute_A, substitute_B, substitution_locations
):
m = 3
stop = 16
T_A_sub = T_A.copy()
T_B_sub = T_B.copy()[:stop]
for substitution_location_B in substitution_locations:
for substitution_location_A in substitution_locations:
T_A_sub[:] = T_A
T_B_sub[:] = T_B[:stop]
T_A_sub[substitution_location_A] = substitute_A
T_B_sub[substitution_location_B] = substitute_B
left = naive.stamp(T_A_sub, m, T_B=T_B_sub)
right = gpu_stump(T_A_sub, m, T_B_sub, ignore_trivial=False)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(
pd.Series(T_A_sub), m, pd.Series(T_B_sub), ignore_trivial=False
)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
def test_gpu_stump_two_constant_subsequences_A_B_join():
T_A = np.array([0, 0, 0, 0, 0, 1], dtype=np.float64)
T_B = np.concatenate(
(np.zeros(20, dtype=np.float64), np.ones(5, dtype=np.float64)))
m = 3
left = naive.stamp(T_A, m, T_B=T_B)
right = gpu_stump(T_A, m, T_B, ignore_trivial=False)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0]) # ignore indices
right = gpu_stump(pd.Series(T_A), m, pd.Series(T_B), ignore_trivial=False)
naive.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0]) # ignore indices
# Swap inputs
left = naive.stamp(T_B, m, T_B=T_A)
right = gpu_stump(T_B, m, T_A, ignore_trivial=False)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0]) # ignore indices
right = gpu_stump(pd.Series(T_B), m, pd.Series(T_A), ignore_trivial=False)
naive.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0]) # ignore indices
def test_parallel_gpu_stump_A_B_join(T_A, T_B):
device_ids = [device.id for device in cuda.list_devices()]
if len(T_B) > 10:
m = 3
left = naive.stamp(T_A, m, T_B=T_B)
right = gpu_stump(
T_A,
m,
T_B,
ignore_trivial=False,
device_id=device_ids,
)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(
pd.Series(T_A),
m,
pd.Series(T_B),
ignore_trivial=False,
device_id=device_ids,
)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
def test_gpu_stump_identical_subsequence_A_B_join():
identical = np.random.rand(8)
T_A = np.random.rand(20)
T_B = np.random.rand(20)
T_A[1 : 1 + identical.shape[0]] = identical
T_B[11 : 11 + identical.shape[0]] = identical
m = 3
left = naive.stamp(T_A, m, T_B=T_B)
right = gpu_stump(T_A, m, T_B, ignore_trivial=False)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(
left[:, 0], right[:, 0], decimal=config.STUMPY_TEST_PRECISION
) # ignore indices
right = gpu_stump(pd.Series(T_A), m, pd.Series(T_B), ignore_trivial=False)
naive.replace_inf(right)
npt.assert_almost_equal(
left[:, 0], right[:, 0], decimal=config.STUMPY_TEST_PRECISION
) # ignore indices
# Swap inputs
left = naive.stamp(T_B, m, T_B=T_A)
right = gpu_stump(T_B, m, T_A, ignore_trivial=False)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(
left[:, 0], right[:, 0], decimal=config.STUMPY_TEST_PRECISION
) # ignore indices
right = gpu_stump(pd.Series(T_B), m, pd.Series(T_A), ignore_trivial=False)
naive.replace_inf(right)
npt.assert_almost_equal(
left[:, 0], right[:, 0], decimal=config.STUMPY_TEST_PRECISION
) # ignore indices
def test_gpu_stump_A_B_join(T_A, T_B):
m = 3
left = naive.stamp(T_A, m, T_B=T_B)
right = gpu_stump(T_A, m, T_B, ignore_trivial=False)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(pd.Series(T_A), m, pd.Series(T_B), ignore_trivial=False)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
def test_gpu_stump_self_join(T_A, T_B):
m = 3
zone = int(np.ceil(m / 4))
ref_mp = naive.stamp(T_B, m, exclusion_zone=zone)
comp_mp = gpu_stump(T_B, m, ignore_trivial=True)
naive.replace_inf(ref_mp)
naive.replace_inf(comp_mp)
npt.assert_almost_equal(ref_mp, comp_mp)
comp_mp = gpu_stump(pd.Series(T_B), m, ignore_trivial=True)
naive.replace_inf(comp_mp)
npt.assert_almost_equal(ref_mp, comp_mp)
def test_gpu_stump_self_join(T_A, T_B):
m = 3
zone = int(np.ceil(m / 4))
left = naive.stamp(T_B, m, exclusion_zone=zone)
right = gpu_stump(T_B, m, ignore_trivial=True)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(pd.Series(T_B), m, ignore_trivial=True)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
def test_stump_A_B_join(T_A, T_B):
m = 3
left = np.array(
[naive_mass(Q, T_A, m) for Q in core.rolling_window(T_B, m)],
dtype=object)
right = gpu_stump(T_A, m, T_B, ignore_trivial=False)
replace_inf(left)
replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(pd.Series(T_A), m, pd.Series(T_B), ignore_trivial=False)
replace_inf(right)
npt.assert_almost_equal(left, right)
def test_gpu_stump_constant_subsequence_self_join():
T_A = np.concatenate((np.zeros(20, dtype=np.float64), np.ones(5, dtype=np.float64)))
m = 3
zone = int(np.ceil(m / 4))
left = naive.stamp(T_A, m, exclusion_zone=zone)
right = gpu_stump(T_A, m, ignore_trivial=True)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0]) # ignore indices
right = gpu_stump(pd.Series(T_A), m, ignore_trivial=True)
naive.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0]) # ignore indices
def test_gpu_stump_self_join_larger_window(T_A, T_B):
for m in [8, 16, 32]:
if len(T_B) > m:
zone = int(np.ceil(m / 4))
left = naive.stamp(T_B, m, exclusion_zone=zone)
right = gpu_stump(T_B, m, ignore_trivial=True)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(pd.Series(T_B), m, ignore_trivial=True,)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
def test_parallel_gpu_stump_self_join(T_A, T_B):
device_ids = [device.id for device in cuda.list_devices()]
if len(T_B) > 10:
m = 3
zone = int(np.ceil(m / 4))
left = naive.stamp(T_B, m, exclusion_zone=zone)
right = gpu_stump(T_B, m, ignore_trivial=True, device_id=device_ids,)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(pd.Series(T_B), m, ignore_trivial=True, device_id=device_ids,)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
def calculate_matrix_profile(column, seq_length):
import stumpy
try:
# stumpy needs np float
old_data = np.array(column, dtype=np.floating)
except ValueError:
raise Exception('Can\'t convert column to float')
try:
if cuda.is_available():
gpu_device_ids = [device.id for device in cuda.list_devices()]
mp = stumpy.gpu_stump(old_data,
m=seq_length,
ignore_trivial=False,
device_id=gpu_device_ids)
else:
mp = stumpy.stump(old_data, m=seq_length, ignore_trivial=False)
except TypeError as e:
print('Type issue in stumpy:')
raise e
except ValueError as e:
print('Seq_length issue in stumpy')
raise e
if pd.isnull(mp).any():
raise Exception(
'Matrix profile for the column contains NaN values. Try to increase the dataset size'
)
return mp
def test_gpu_stump_A_B_join(T_A, T_B):
m = 3
ref_mp = naive.stamp(T_B, m, T_B=T_A)
comp_mp = gpu_stump(T_B, m, T_A, ignore_trivial=False)
naive.replace_inf(ref_mp)
naive.replace_inf(comp_mp)
npt.assert_almost_equal(ref_mp, comp_mp)
def test_gpu_stump_identical_subsequence_self_join():
identical = np.random.rand(8)
T_A = np.random.rand(20)
T_A[1:1 + identical.shape[0]] = identical
T_A[11:11 + identical.shape[0]] = identical
m = 3
zone = int(np.ceil(m / 4))
left = naive.stamp(T_A, m, exclusion_zone=zone)
right = gpu_stump(T_A, m, ignore_trivial=True)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0],
decimal=naive.PRECISION) # ignore indices
right = gpu_stump(pd.Series(T_A), m, ignore_trivial=True)
naive.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0],
decimal=naive.PRECISION) # ignore indices
def test_stump_self_join(T_A, T_B):
m = 3
zone = int(np.ceil(m / 4))
left = np.array(
[
naive_mass(Q, T_B, m, i, zone, True)
for i, Q in enumerate(core.rolling_window(T_B, m))
],
dtype=object,
)
right = gpu_stump(T_B, m, ignore_trivial=True)
replace_inf(left)
replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(pd.Series(T_B), m, ignore_trivial=True)
replace_inf(right)
npt.assert_almost_equal(left, right)
def test_gpu_stump_self_join_larger_window(T_A, T_B, m):
if len(T_B) > m:
zone = int(np.ceil(m / 4))
ref_mp = naive.stamp(T_B, m, exclusion_zone=zone)
comp_mp = gpu_stump(T_B, m, ignore_trivial=True)
naive.replace_inf(ref_mp)
naive.replace_inf(comp_mp)
npt.assert_almost_equal(ref_mp, comp_mp)
def test_gpu_stump_constant_subsequence_self_join():
T_A = np.concatenate((np.zeros(20, dtype=np.float64), np.ones(5, dtype=np.float64)))
m = 3
zone = int(np.ceil(m / 4))
ref_mp = naive.stamp(T_A, m, exclusion_zone=zone)
comp_mp = gpu_stump(T_A, m, ignore_trivial=True)
naive.replace_inf(ref_mp)
naive.replace_inf(comp_mp)
npt.assert_almost_equal(ref_mp[:, 0], comp_mp[:, 0]) # ignore indices
def _transform(self, X, y=None):
n_ts, sz, d = X.shape
if d > 1:
raise NotImplementedError("We currently don't support using "
"multi-dimensional matrix profiles "
"from the stumpy library.")
output_size = sz - self.subsequence_length + 1
X_transformed = np.empty((n_ts, output_size, 1))
if self.implementation == "stump":
if not STUMPY_INSTALLED:
raise ImportError(stumpy_msg)
for i_ts in range(n_ts):
result = stumpy.stump(T_A=X[i_ts, :, 0].ravel(),
m=self.subsequence_length)
X_transformed[i_ts, :, 0] = result[:, 0].astype(np.float)
elif self.implementation == "gpu_stump":
if not STUMPY_INSTALLED:
raise ImportError(stumpy_msg)
for i_ts in range(n_ts):
result = stumpy.gpu_stump(T_A=X[i_ts, :, 0].ravel(),
m=self.subsequence_length)
X_transformed[i_ts, :, 0] = result[:, 0].astype(np.float)
elif self.implementation == "numpy":
scaler = TimeSeriesScalerMeanVariance()
band_width = int(np.ceil(self.subsequence_length / 4))
for i_ts in range(n_ts):
segments = _series_to_segments(X[i_ts],
self.subsequence_length)
if self.scale:
segments = scaler.fit_transform(segments)
n_segments = segments.shape[0]
segments_2d = segments.reshape(
(-1, self.subsequence_length * d))
dists = squareform(pdist(segments_2d, "euclidean"))
band = (np.tri(
n_segments, n_segments, band_width, dtype=np.bool
) & ~np.tri(
n_segments, n_segments, -(band_width + 1), dtype=np.bool))
dists[band] = np.inf
X_transformed[i_ts] = dists.min(axis=1, keepdims=True)
else:
available_implementations = ["numpy", "stump", "gpu_stump"]
raise ValueError(
'This "{}" matrix profile implementation is not'
' recognized. Available implementations are {}.'.format(
self.implementation, available_implementations))
return X_transformed
def test_gpu_stump_nan_inf_self_join(T_A, T_B, substitute_B, substitution_locations):
m = 3
stop = 16
T_B_sub = T_B.copy()[:stop]
for substitution_location_B in substitution_locations:
T_B_sub[:] = T_B[:stop]
T_B_sub[substitution_location_B] = substitute_B
zone = int(np.ceil(m / 4))
left = naive.stamp(T_B_sub, m, exclusion_zone=zone)
right = gpu_stump(T_B_sub, m, ignore_trivial=True)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(pd.Series(T_B_sub), m, ignore_trivial=True)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
def test_gpu_stump(T, m):
if not cuda.is_available(): # pragma: no cover
pytest.skip("Skipping Tests No GPUs Available")
if T.ndim > 1:
T = T.copy()
T = T[0]
ref = stumpy.gpu_aamp(T, m)
comp = stumpy.gpu_stump(T, m, normalize=False)
npt.assert_almost_equal(ref, comp)
def test_constant_subsequence_self_join():
T_A = np.concatenate(
(np.zeros(20, dtype=np.float64), np.ones(5, dtype=np.float64)))
m = 3
zone = int(np.ceil(m / 4))
left = np.array(
[
utils.naive_mass(Q, T_A, m, i, zone, True)
for i, Q in enumerate(core.rolling_window(T_A, m))
],
dtype=object,
)
right = gpu_stump(T_A, m, ignore_trivial=True)
utils.replace_inf(left)
utils.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0]) # ignore indices
right = gpu_stump(pd.Series(T_A), m, ignore_trivial=True)
utils.replace_inf(right)
npt.assert_almost_equal(left[:, 0], right[:, 0]) # ignore indices
def test_gpu_stump_one_constant_subsequence_A_B_join():
T_A = np.random.rand(20)
T_B = np.concatenate((np.zeros(20, dtype=np.float64), np.ones(5, dtype=np.float64)))
m = 3
ref_mp = naive.stamp(T_B, m, T_B=T_A)
comp_mp = gpu_stump(T_B, m, T_A, ignore_trivial=False)
naive.replace_inf(ref_mp)
naive.replace_inf(comp_mp)
npt.assert_almost_equal(ref_mp[:, 0], comp_mp[:, 0]) # ignore indices
# comp_mp = gpu_stump(pd.Series(T_A), m, pd.Series(T_B), ignore_trivial=False)
# naive.replace_inf(comp_mp)
# npt.assert_almost_equal(ref_mp[:, 0], comp_mp[:, 0]) # ignore indices
# Swap inputs
ref_mp = naive.stamp(T_A, m, T_B=T_B)
comp_mp = gpu_stump(T_A, m, T_B, ignore_trivial=False)
naive.replace_inf(ref_mp)
naive.replace_inf(comp_mp)
npt.assert_almost_equal(ref_mp[:, 0], comp_mp[:, 0]) # ignore indices
def test_gpu_stump_nan_zero_mean_self_join():
T = np.array([-1, 0, 1, np.inf, 1, 0, -1])
m = 3
zone = int(np.ceil(m / 4))
left = naive.stamp(T, m, exclusion_zone=zone)
right = gpu_stump(T, m, ignore_trivial=True)
naive.replace_inf(left)
naive.replace_inf(right)
npt.assert_almost_equal(left, right)
def test_gpu_stump_A_B_join(T_A, T_B):
m = 3
left = np.array(
[utils.naive_mass(Q, T_A, m) for Q in core.rolling_window(T_B, m)],
dtype=object)
right = gpu_stump(T_A,
m,
T_B,
ignore_trivial=False,
threads_per_block=THREADS_PER_BLOCK)
utils.replace_inf(left)
utils.replace_inf(right)
npt.assert_almost_equal(left, right)
right = gpu_stump(
pd.Series(T_A),
m,
pd.Series(T_B),
ignore_trivial=False,
threads_per_block=THREADS_PER_BLOCK,
)
utils.replace_inf(right)
npt.assert_almost_equal(left, right)
def test_gpu_stump_identical_subsequence_self_join():
identical = np.random.rand(8)
T_A = np.random.rand(20)
T_A[1:1 + identical.shape[0]] = identical
T_A[11:11 + identical.shape[0]] = identical
m = 3
zone = int(np.ceil(m / 4))
ref_mp = naive.stamp(T_A, m, exclusion_zone=zone)
comp_mp = gpu_stump(T_A, m, ignore_trivial=True)
naive.replace_inf(ref_mp)
naive.replace_inf(comp_mp)
npt.assert_almost_equal(
ref_mp[:, 0], comp_mp[:, 0],
decimal=config.STUMPY_TEST_PRECISION) # ignore indices
