def preprocess_X_y_FeaExtr(ts, X, y):
new_X = []
new_y = []
winlength_seconds = 1
overlap_seconds = 0
print("Window size of " + str(winlength_seconds) +
" seconds and overlap of" +
str(float(overlap_seconds) * 100 / winlength_seconds) + "%")
samples_per_sec = 5
winlength = samples_per_sec * winlength_seconds
current = winlength
overlap = samples_per_sec * overlap_seconds
transform = Transforms(window_length=winlength, window_overlap=overlap)
print("Use number of mean crossings, spectral entropy as features...")
feature_transforms = [
transform.mean, transform.std, transform.prod, transform.minx,
transform.maxx
]
# transform.sumx]
# transform.mean_crossings]
# transform.spec_entropy,
# transform.zero_crossings]
# transform.interq,
# transform.skewn,
# transform.spec_energy,
# transform.p25,
# transform.p75,
# transform.kurtosis]
while True:
windowed_raw = transform.slide(X)
if len(windowed_raw) > 0:
try:
# TODO Save the timestamps into a windowed_ts instead
windowed_features = [ts[transform.current_position][0]]
except Exception as e:
print(e)
break
for function in feature_transforms:
windowed_features.extend(
(np.apply_along_axis(function, 0, windowed_raw).tolist()))
new_X.append(windowed_features)
windowed_raw_labels = transform.slide(y, update=False)
most_freq_label = np.bincount(windowed_raw_labels).argmax()
new_y.append(most_freq_label)
# Convert lists to Numpy arrays
new_X = np.array(new_X)
# Store new timestamps
# TODO This will not be necessary after the windowed_ts is generated
ts = new_X[:, 0]
# TODO Check why the features are converted to strings
# Remove datetime from features
new_X = new_X[:, 1:].astype(float)
new_y = np.array(new_y)
return ts, new_X, new_y
def test_spec_entropy(self):
"""
Test spectral entropy.
"""
print('Test spectral entropy.')
x_list = [[5, 6, 7, 8, 9, 10], [1, 2, 3, 4, 2, 3, 4, 10, 5, 7, 8, 11],
[7, 7, 31, 31, 47, 75, 87, 115, 116, 119, 119, 155, 177]]
y_list = [0.9333118993685063, 1.8492364375192922, 1.941214222324683]
t = Transforms(window_length=1, stride=0)
for x, y in zip(x_list, y_list):
result = t.spec_entropy(x)
np.testing.assert_almost_equal(result, y)
def test_spec_energy(self):
"""
Test spectral energy.
"""
print('Test spectral energy.')
x_list = [[5, 6, 7, 8, 9, 10], [1, 2, 3, 4, 2, 3, 4, 10, 5, 7, 8, 11],
[7, 7, 31, 31, 47, 75, 87, 115, 116, 119, 119, 155, 177]]
y_list = [2130, 5016, 1661139.9999999995]
t = Transforms(window_length=1, stride=0)
for x, y in zip(x_list, y_list):
result = t.spec_energy(x)
np.testing.assert_almost_equal(result, y)
def preprocess_X_y_Shape(ts, X, y):
new_X = []
new_y = []
winlength_seconds = 10
overlap_seconds = 9
print("Window size of " + str(winlength_seconds) +
" seconds and overlap of" +
str(float(overlap_seconds) * 100 / winlength_seconds) + "%")
samples_per_sec = 1
winlength = samples_per_sec * winlength_seconds
current = winlength
overlap = samples_per_sec * overlap_seconds
transform = Transforms(window_length=winlength, window_overlap=overlap)
print("Use number of mean crossings, spectral entropy as features...")
feature_transforms = [transform.raw]
while True:
windowed_raw = transform.slide(X)
if len(windowed_raw) > 0:
try:
windowed_features = [ts[transform.current_position][0]]
except Exception as e:
print(e)
break
for function in feature_transforms:
windowed_features.extend(
(np.apply_along_axis(function, 0, windowed_raw).tolist()))
new_X.append(windowed_features)
windowed_raw_labels = transform.slide(y, update=False)
most_freq_label = np.bincount(windowed_raw_labels).argmax()
new_y.append(most_freq_label)
# Convert lists to Numpy arrays
new_X = np.array(new_X)
# Store new timestamps
ts = new_X[:, 0]
# Remove datetime from features
new_X = new_X[:, 1:]
new_y = np.array(new_y)
dims = np.size(new_X[0][0])
tmpX = np.ones((new_X.shape[0], new_X.shape[1], dims), dtype='float64')
for i in range(0, new_X.shape[0]):
for j in range(0, new_X.shape[1]):
for k in range(0, dims):
tmpX[i][j][k] = new_X[i][j][k]
new_X = tmpX
return ts, new_X, new_y
def test_25_percentiles(self):
"""
Test 25th percentiles.
"""
print('Test 25th percentiles.')
x_list = [[1, 2, 3, 4, 5, 6, 7], [5, 6, 3, 1, 3],
[10, -10, -90, -90.5, -100, -200]]
y_list = [2.5, 3, -97.625]
t = Transforms(window_length=1, stride=0)
for x, y in zip(x_list, y_list):
result = t.p25(x)
self.assertEqual(result, y)
winlength = 3
x_list = [[1, 2, 3, 4, 5, 6, 7], [5, 6, 3, 1, 3],
[10, -10, -90, -90.5, -100, -200]]
y_list = [[1.5, 2.5, 3.5, 4.5, 5.5], [4, 2, 2],
[-50, -90.25, -95.25, -150]]
t = Transforms(window_length=winlength, stride=1)
for x, y in zip(x_list, y_list):
t.current_position = winlength
final_result = []
while True:
windowed_raw = t.slide(np.array(x))
if len(windowed_raw) > 0 and len(windowed_raw) == winlength:
result = t.p25(windowed_raw)
final_result.append(result)
else:
break
np.testing.assert_almost_equal(final_result, y)
def test_zero_crossings(self):
"""
Test zero crossings.
"""
print('Test zero crossings.')
x_list = [[-1, -1, -1], [0, 0, 0], [1, -4, 5, -10], [2, -3, 4, -5]]
y_list = [0, 0, 3, 3]
t = Transforms(window_length=1, stride=0)
for x, y in zip(x_list, y_list):
result = t.zero_crossings(x)
self.assertEqual(result, y)
winlength = 3
x_list = [[-1, -1, -1, -1, -1, -1], [0, 0, 0, 0, 0, 0],
[1, -5, 2, -6, 9, -3, 2, -5, 2], [-1, 1, 2, 2, -3, 3]]
y_list = [[0, 0, 0, 0], [0, 0, 0, 0], [2, 2, 2, 2, 2, 2, 2],
[1, 0, 1, 2]]
t = Transforms(window_length=winlength, stride=1)
for x, y in zip(x_list, y_list):
t.current_position = winlength
final_result = []
while True:
windowed_raw = t.slide(np.array(x))
if len(windowed_raw) > 0 and len(windowed_raw) == winlength:
result = t.zero_crossings(windowed_raw)
final_result.append(result)
else:
break
self.assertEqual(final_result, y)
def test_kurtosis(self):
"""
Test kurtosis.
"""
print('Test kurtosis.')
x_list = [[5, 6, 7, 8, 9, 10], [1, 2, 3, 4, 2, 3, 4, 10, 5, 7, 8, 11],
[7, 7, 31, 31, 47, 75, 87, 115, 116, 119, 119, 155, 177]]
y_list = [1.7314285714285718, 2.128698649813272, 1.8298588697064042]
t = Transforms(window_length=1, stride=0)
for x, y in zip(x_list, y_list):
result = t.kurtosis(x)
self.assertEqual(result, y)
winlength = 5
x_list = [[1, 2, 3, 4, 5, 6, 7], [5, 6, 3, 1, 3],
[10, -90, -10, -90.5, -100, -200]]
y_list = [[1.7, 1.7, 1.7], [1.7957064], [1.2987871, 2.5169668]]
t = Transforms(window_length=winlength, stride=1)
for x, y in zip(x_list, y_list):
t.current_position = winlength
final_result = []
while True:
windowed_raw = t.slide(np.array(x))
if len(windowed_raw) > 0 and len(windowed_raw) == winlength:
result = t.kurtosis(windowed_raw)
final_result.append(result)
else:
break
np.testing.assert_almost_equal(final_result, y)
def test_skewness(self):
"""
Test skewness.
"""
print('Test skewness.')
x_list = [[5, 6, 7, 8, 9, 10], [1, 2, 3, 4, 2, 3, 4, 10, 5, 7, 8, 11],
[7, 7, 31, 31, 47, 75, 87, 115, 116, 119, 119, 155, 177]]
y_list = [0, 0.6486028516711329, 0.05039822936776113]
t = Transforms(window_length=1, stride=0)
for x, y in zip(x_list, y_list):
result = t.skewn(x)
np.testing.assert_almost_equal(result, y)
winlength = 3
x_list = [[1, 2, 3, 4, 5, 6, 7], [5, 6, 3, 1, 3],
[10, -90, -10, -90.5, -100, -200]]
y_list = [[0, 0, 0, 0, 0], [-0.3818018, 0.2390631, -0.7071068],
[-0.5951701, 0.7070141, 0.6778576, -0.687648]]
t = Transforms(window_length=winlength, stride=1)
for x, y in zip(x_list, y_list):
t.current_position = winlength
final_result = []
while True:
windowed_raw = t.slide(np.array(x))
if len(windowed_raw) > 0 and len(windowed_raw) == winlength:
result = t.skewn(windowed_raw)
final_result.append(result)
else:
break
np.testing.assert_almost_equal(final_result, y)
def test_interquartiles(self):
"""
Test interquartiles.
"""
print('Test interquartiles.')
x_list = [[5, 6, 7, 8, 9, 10], [1, 2, 3, 4, 2, 3, 4, 10, 5, 7, 8, 11],
[7, 7, 31, 31, 47, 75, 87, 115, 116, 119, 119, 155, 177]]
y_list = [2.5, 4.5, 88]
t = Transforms(window_length=1, stride=0)
for x, y in zip(x_list, y_list):
result = t.interq(x)
self.assertEqual(result, y)
winlength = 3
x_list = [[1, 2, 3, 4, 5, 6, 7], [5, 6, 3, 1, 3],
[10, -90, -10, -90.5, -100, -200]]
y_list = [[1, 1, 1, 1, 1], [1.5, 2.5, 1], [50, 40.25, 45, 54.75]]
t = Transforms(window_length=winlength, stride=1)
for x, y in zip(x_list, y_list):
t.current_position = winlength
final_result = []
while True:
windowed_raw = t.slide(np.array(x))
if len(windowed_raw) > 0 and len(windowed_raw) == winlength:
result = t.interq(windowed_raw)
final_result.append(result)
else:
break
np.testing.assert_almost_equal(final_result, y)
def preprocess_X_y(ts, X, y):
new_X = []
new_y = []
winlength_seconds = 5
overlap_seconds = 1
print("Window size of " + str(winlength_seconds) +
" seconds and overlap of " +
str(float(overlap_seconds) / winlength_seconds) + "%")
samples_per_sec = 50
winlength = samples_per_sec * winlength_seconds
current = winlength
overlap = samples_per_sec * overlap_seconds
transform = Transforms(window_length=winlength, window_overlap=overlap)
print("Use number of mean crossings, spectral entropy as features...")
feature_transforms = [
transform.mean_crossings, transform.spec_entropy,
transform.zero_crossings, transform.interq, transform.skewn,
transform.spec_energy, transform.p25, transform.p75, transform.kurtosis
]
while True:
try:
ts_window = ts[transform.current_position - 1]
except Exception as e:
print(e)
break
windowed_raw = transform.slide(X, update=False)
windowed_raw_labels = transform.slide(y, update=True)
if len(windowed_raw) > 0:
windowed_features = [ts_window]
for function in feature_transforms:
windowed_features.extend(
(np.apply_along_axis(function, 0, windowed_raw).tolist()))
new_X.append(windowed_features)
most_freq_label = np.bincount(windowed_raw_labels).argmax()
new_y.append(most_freq_label)
# Convert lists to Numpy arrays
new_X = np.array(new_X)
# Remove datetime from features
new_X = new_X[:, 1:]
new_y = np.array(new_y)
new_X = transform.feature_selection(new_X, new_y, 'tree')
return new_X, new_y
def preprocess_X_y_2(ts, X, y):
new_X = []
new_y = []
winlength_seconds = 10
overlap_seconds = 5
print("Window size of " + str(winlength_seconds) +
" seconds and overlap of" +
str(float(overlap_seconds) * 100 / winlength_seconds) + "%")
samples_per_sec = 1
winlength = samples_per_sec * winlength_seconds
current = winlength
overlap = samples_per_sec * overlap_seconds
transform = Transforms(window_length=winlength, window_overlap=overlap)
print("Use number of mean crossings, spectral entropy as features...")
feature_transforms = [transform.mean]
from IPython import embed
embed()
while True:
windowed_raw = transform.slide(X)
if len(windowed_raw) > 0:
try:
windowed_features = [ts[transform.current_position][0]]
except Exception as e:
print(e)
break
for function in feature_transforms:
windowed_features.extend(
(np.apply_along_axis(function, 0, windowed_raw).tolist()))
new_X.append(windowed_features)
windowed_raw_labels = transform.slide(y, update=False)
most_freq_label = np.bincount(windowed_raw_labels).argmax()
new_y.append(most_freq_label)
# Convert lists to Numpy arrays
new_X = np.array(new_X)
# Store new timestamps
ts = new_X[:, 0]
# Remove datetime from features
new_X = new_X[:, 1:]
new_y = np.array(new_y)
return ts, new_X, new_y