def test_index_mass_quantile_relative(x):
result = IndexMassQuantile(q=0.5, rel=True).transform(x)
for i, a in series(x):
actual = result.values[i]
a = np.abs(a)
expected = (np.argmax((np.cumsum(a) / np.sum(a)) >= 0.5) + 1) / len(a)
np.testing.assert_almost_equal(actual, expected)
def test_entropy(x):
x = Collection.from_array(np.abs(x.values) + 1) # ensure that values > 0
result = Entropy().transform(x)
for i, a in series(x):
actual = result.values[i]
expected = stats.entropy(a)
np.testing.assert_almost_equal(actual, expected)
def test_ratio_absolute(x, y):
result = Ratio().transform(x, y)
for i, a, b in series(x, y):
actual = result.values[i]
with np.errstate(divide='ignore', invalid='ignore'):
expected = a / b
np.testing.assert_almost_equal(actual, expected)
def test_high_standard_deviation_one(x):
result = HighStandardDeviation(r=1).transform(x)
assert np.issubdtype(result.dtype, np.bool_)
for i, a in series(x):
actual = result.values[i]
expected = np.std(a) > 1 * (np.max(a) - np.min(a))
np.testing.assert_almost_equal(actual, expected)
def test_number_of_peaks_cwt(x):
result = NumberPeaksCWT(max_width=5).transform(x)
for i, a in series(x):
actual = result.values[i]
expected = len(
signal.find_peaks_cwt(a, np.arange(1, 6), wavelet=signal.ricker))
np.testing.assert_almost_equal(actual, expected)
def test_high_variance_1(x):
result = HighVariance(threshold=1).transform(x)
assert np.issubdtype(result.dtype, np.bool_)
for i, a in series(x):
actual = result.values[i]
expected = np.var(a) > 1
np.testing.assert_almost_equal(actual, expected)
def test_difference_relative(x, y):
result = Difference(rel=True).transform(x, y)
for i, a, b in series(x, y):
actual = result.values[i]
with np.errstate(divide='ignore', invalid='ignore'):
expected = np.abs((b - a) / a)
np.testing.assert_almost_equal(actual, expected)
def test_resample_down(x):
num = int(x.shape[1] / 2)
result = Resample(num=num).transform(x)
for i, a in series(x):
actual = result.values[i]
expected = signal.resample(a, num=num)
np.testing.assert_almost_equal(actual, expected)
def test_time_reversal_asymmetry_statistic(x):
result = TimeReversalAsymmetryStatistic(lag=1).transform(x)
for i, a in series(x):
actual = result.values[i]
n = len(a) - 2
expected = np.mean(a[2:][:n] * a[2:][:n] * a[1:][:n] - a[1:][:n] * a[:n] * a[:n])
np.testing.assert_almost_equal(actual, expected)
def test_cross_correlation(x, y):
result = CrossCorrelation().transform(x, y)
for i, a, b in series(x, y):
actual = result.values[i]
np.testing.assert_almost_equal(actual[0], np.correlate(a, b)[0])
for lag in range(1, len(actual)):
np.testing.assert_almost_equal(actual[lag], np.correlate(a[lag:], b[:-lag])[0])
def test_symmetry_looking_1(x):
result = SymmetryLooking(r=1).transform(x)
assert np.issubdtype(result.dtype, np.bool_)
for i, a in series(x):
actual = result.values[i]
expected = np.abs(np.mean(a) - np.median(a)) < 1 * (np.max(a) - np.min(a))
np.testing.assert_almost_equal(actual, expected)
def test_fft_moments(x):
def moment(a, moment):
return a.dot(np.arange(len(a))**moment) / np.sum(a)
def mean(a):
return moment(a, 1)
def variance(a):
return moment(a, 2) - mean(a)**2
def skewness(a):
return (moment(a, 3) - 3 * mean(a) * variance(a) -
mean(a)**3) / variance(a)**(1.5)
def kurtosis(a):
return ((moment(a, 4) - 4 * mean(a) * moment(a, 3) +
6 * moment(a, 2) * mean(a)**2 - 3 * mean(a)) / variance(a)**2)
fft = FFT().transform(x)
result_mean = SpectralMean().transform(x)
result_variance = SpectralVariance().transform(x)
result_skewness = SpectralSkewness().transform(x)
result_kurtosis = SpectralKurtosis().transform(x)
with warnings.catch_warnings():
warnings.simplefilter("ignore")
for i, a in series(x):
np.testing.assert_allclose(result_mean.values[i], mean(a))
np.testing.assert_allclose(result_variance.values[i], variance(a))
np.testing.assert_allclose(result_skewness.values[i], skewness(a))
np.testing.assert_allclose(result_kurtosis.values[i], kurtosis(a))
def test_auto_correlation(x):
result = AutoCorrelation().transform(x)
for i, a in series(x):
actual = result.values[i]
mu = np.mean(a)
var = np.var(a)
np.testing.assert_almost_equal(actual[0], np.mean((a - mu) * (a - mu)) / var)
for lag in range(1, len(actual)):
expected = np.mean((a[:-lag] - mu) * (a[lag:] - mu)) / var
np.testing.assert_almost_equal(actual[lag], expected)
def test_friedrich_coefficients_m_3(x):
result = FriedrichCoefficients(m=3, r=10).transform(x)
for i, a in series(x):
actual = result.values[i]
df = pd.DataFrame({'signal': a[:-1], 'delta': np.diff(a)})
df['quantiles'] = pd.qcut(df.signal, 10)
q = df.groupby('quantiles')
mean = pd.DataFrame({'x': q.signal.mean(), 'y': q.delta.mean()})
expected = np.polyfit(mean.x, mean.y, deg=3)
np.testing.assert_almost_equal(actual, expected)
def test_max_langevin_fixed_point_m_3(x):
result = MaxLangevinFixedPoint(m=3, r=10).transform(x)
for i, a in series(x):
actual = result.values[i]
df = pd.DataFrame({'signal': a[:-1], 'delta': np.diff(a)})
df['quantiles'] = pd.qcut(df.signal, 10)
q = df.groupby('quantiles')
mean = pd.DataFrame({'x': q.signal.mean(), 'y': q.delta.mean()})
coef = np.polyfit(mean.x, mean.y, deg=3)
expected = np.max(np.roots(coef).real)
np.testing.assert_almost_equal(actual, expected)
def test_single_pass_statistics(x):
result = SinglePassStatistics().transform(x)
for i, a in series(x):
actual = result.values[i]
np.testing.assert_almost_equal(actual[0], len(a))
np.testing.assert_almost_equal(actual[1], np.sum(a))
np.testing.assert_almost_equal(actual[2], np.min(a))
np.testing.assert_almost_equal(actual[3], np.max(a))
np.testing.assert_almost_equal(actual[4], np.mean(a))
np.testing.assert_almost_equal(actual[5], np.var(a))
np.testing.assert_almost_equal(actual[6], stats.skew(a))
np.testing.assert_almost_equal(actual[7], stats.kurtosis(a))
def test_binned_distribution(x):
result = BinnedDistribution(bins=10).transform(x)
for i, a in series(x):
actual = result.values[i]
r = (np.min(a), np.max(a))
s = (r[1] - r[0]) / 10
for bin in range(10):
lower = r[0] + bin * s
upper = r[0] + (bin + 1) * s
if bin < 10 - 1:
count = len(np.where((a >= lower) & (a < upper))[0])
else:
count = len(np.where(a >= lower)[0])
ratio = count / len(a)
np.testing.assert_almost_equal(actual[bin], ratio)
def test_binned_entropy(x):
result = BinnedEntropy(bins=10).transform(x)
for i, a in series(x):
actual = result.values[i]
r = (np.min(a), np.max(a))
s = (r[1] - r[0]) / 10
bd = []
for bin in range(10):
lower = r[0] + bin * s
upper = r[0] + (bin + 1) * s
if bin < 10 - 1:
count = len(np.where((a >= lower) & (a < upper))[0])
else:
count = len(np.where(a >= lower)[0])
bd.append(count / len(a))
np.testing.assert_almost_equal(actual, stats.entropy(bd))
def test_linear_trend(x):
result = LinearTrend().transform(x)
for i, a in series(x):
actual = result.values[i]
expected = stats.linregress(np.arange(len(a)), a)
np.testing.assert_almost_equal(actual, expected)
def test_count_below_mean(x):
result = CountBelowMean().transform(x)
for i, a in series(x):
actual = result.values[i]
expected = np.sum(a < np.mean(a))
np.testing.assert_almost_equal(actual, expected)
def test_kurtosis(x):
result = Kurtosis().transform(x)
for i, a in series(x):
actual = result.values[i]
expected = stats.kurtosis(a)
np.testing.assert_almost_equal(actual, expected)
def test_skewness(x):
result = Skewness().transform(x)
for i, a in series(x):
actual = result.values[i]
expected = stats.skew(a)
np.testing.assert_almost_equal(actual, expected)
def test_variance(x):
result = Variance().transform(x)
for i, a in series(x):
actual = result.values[i]
expected = np.var(a)
np.testing.assert_almost_equal(actual, expected)
def test_median(x):
result = Median().transform(x)
for i, a in series(x):
actual = result.values[i]
expected = np.median(a)
np.testing.assert_almost_equal(actual, expected)
def test_range_count(x):
result = RangeCount(min=0, max=1).transform(x)
for i, a in series(x):
actual = result.values[i]
expected = np.sum((a >= 0) & (a < 1))
np.testing.assert_almost_equal(actual, expected)
def test_outliers_rel(x):
result = Outliers(r=1.5, rel=True).transform(x)
for i, a in series(x):
actual = result.values[i]
expected = np.sum(np.abs(a - np.mean(a)) > 1.5 * np.std(a)) / len(a)
np.testing.assert_almost_equal(actual, expected)
def test_auto_regressive_coefficients(x):
result = AutoRegressiveCoefficients().transform(x)
for i, a in series(x):
actual = result.values[i]
expected = statsmodels.tsa.ar_model.AR(a).fit().params
np.testing.assert_almost_equal(actual, expected)
def test_mean_second_derivative_central(x):
result = MeanSecondDerivativeCentral().transform(x)
for i, a in series(x):
actual = result.values[i]
expected = np.mean(a[2:] - 2 * a[1:-1] + a[:-2])
np.testing.assert_almost_equal(actual, expected)
def test_mean_change_abs(x):
result = MeanChange(abs=True).transform(x)
for i, a in series(x):
actual = result.values[i]
expected = np.mean(np.abs(np.diff(a)))
np.testing.assert_almost_equal(actual, expected)
def test_sum_change(x):
result = SumChange().transform(x)
for i, a in series(x):
actual = result.values[i]
expected = np.sum(np.diff(a))
np.testing.assert_almost_equal(actual, expected)