def objective(std_noise):
profile = mp.stomp(sequence, window_size, std_noise=std_noise)[0]
if np.max(profile) == 0:
return np.inf
return np.mean(profile) / np.max(profile)
def compute_mp(ts, window, threshold=None):
"""
Compute matrix profile at given window
Args:
ts - array containing time series data
window - window length
threshold - threshold for outlier value
Return:
numpy array - matrix profile
"""
# sfp - commenting this out as it is unneccesary
# remove trailing nans of ts
#i = len(ts) - 1
#while np.isnan(ts[i]) and i >= 0:
# i -= 1
#ts = ts[0:i+1]
# compute mp by stamp
mp = np.array(matrixProfile.stomp(
ts,
m=window,
))[0]
# calibrate ts and mp, so mp value is assigned to the middle of that window
mp_head = np.zeros(window // 2 - 1)
mp_tail = np.zeros(len(ts) - len(mp) - window // 2 + 1)
mp = np.concatenate([mp_head, mp, mp_tail])
# remove error results due to zero std (make them 0 so they don't contribute to outliers)
ts_std = compute_std(ts, window=window)
count_zero_std = 0
for i in range(len(ts_std)):
if ts_std[i] == 0:
mp[i] = 0
count_zero_std += 1
# compute percentage of outliers, where head, tail and zero std points do not participate
if not threshold is None:
outlier = mp[np.where(mp > threshold)]
outlier_percentage = len(outlier) / (len(mp) - len(mp_head) -
len(mp_tail) - count_zero_std)
print('outlier %: ' + str(outlier_percentage))
return mp
def test_extract_regimes(self):
data_file = os.path.join(MODULE_PATH, '..', 'docs', 'examples',
'rawdata.csv')
ts = np.loadtxt(data_file, skiprows=1)
m = 32
mp, pi = stomp(ts, m)
cac = fluss(pi, m)
# test with 3 regimes
regimes = extract_regimes(cac, m)
expected = np.array([759, 423, 583])
np.testing.assert_array_equal(regimes, expected)
# test with 2 regimes
regimes = extract_regimes(cac, m, num=2)
expected = np.array([759, 423])
np.testing.assert_array_equal(regimes, expected)
def compute_mp(ts, window, threshold):
# remove trailing nans of ts
i = len(ts) - 1
while np.isnan(ts[i]) and i >= 0:
i -= 1
ts = ts[0:i]
# compute mp by stamp
mp = np.array(matrixProfile.stomp(ts, m=window))[0]
# calibrate ts and mp, so mp value is assigned to the middle of that window
mp_head = np.zeros(window // 2 - 1)
mp_tail = np.zeros(len(ts) - len(mp) - window // 2 + 1)
mp = np.concatenate([mp_head, mp, mp_tail])
# remove error results due to zero std (make them 0 so they don't contribute to outliers)
ts_std = compute_std(ts, window=window)
count_zero_std = 0
for i in range(len(ts_std)):
if ts_std[i] == 0:
mp[i] = 0
count_zero_std += 1
# compute percentage of outliers, where head, tail and zero std points do not participate
outlier = mp[np.where(mp > threshold)]
outlier_percentage = len(outlier) / (len(mp) - len(mp_head) -
len(mp_tail) - count_zero_std)
return mp, outlier_percentage
def matrixprofile(self, col, motiflen):
mp = matrixProfile.stomp(self.data[castlist(col)].values.flatten(), motiflen)
return self.scatter(y = mp[0], name = 'matrixprofile (len {})'.format(motiflen))
#doc_path = '/home/bhossein/BMBF project/code_resources/matrixprofile-ts-master/'
doc_path = 'C:\Hinkelstien\code_resources\matrixprofile-ts-master/'
data = pd.read_csv(doc_path + 'docs/examples/rawdata.csv')
pattern = data.data.values
#Plot data
fig, ax1 = plt.subplots(figsize=(20, 10))
ax1.plot(np.arange(len(pattern)), pattern, label="Synthetic Data")
legend = ax1.legend(loc='upper right')
#%% Calculate the Matrix Profile
m = 32
mp = matrixProfile.stomp(pattern, m)
# Bonus: calculate the corrected arc curve (CAC) to do semantic segmantation.
cac = fluss.fluss(mp[1], m)
#Append np.nan to Matrix profile to enable plotting against raw data
mp_adj = np.append(mp[0], np.zeros(m - 1) + np.nan)
#Plot the signal data
fig, (ax1, ax2, ax3) = plt.subplots(3, 1, sharex=True, figsize=(20, 10))
ax1.plot(np.arange(len(pattern)), pattern, label="Synthetic Data")
ax1.set_ylabel('Signal', size=22)
#Plot the Matrix Profile
ax2.plot(np.arange(len(mp_adj)), mp_adj, label="Matrix Profile", color='red')
ax2.set_ylabel('Matrix Profile', size=22)