def _compute_expected_results(X,
y=None,
n_coefs=None,
n_bins=4,
strategy='quantile',
drop_sum=False,
anova=False,
norm_mean=False,
norm_std=False,
alphabet=None):
"""Compute the expected results."""
X = np.asarray(X)
if norm_mean:
X -= X.mean(axis=1)[:, None]
if norm_std:
X /= X.std(axis=1)[:, None]
X_fft = np.fft.rfft(X)
X_fft = np.vstack([np.real(X_fft), np.imag(X_fft)])
X_fft = X_fft.reshape(n_samples, -1, order='F')
if drop_sum:
X_fft = X_fft[:, 2:-1]
else:
X_fft = np.hstack([X_fft[:, :1], X_fft[:, 2:-1]])
if n_coefs is not None:
if anova:
_, p = f_classif(X_fft, y)
support = np.argsort(p)[:n_coefs]
X_fft = X_fft[:, support]
else:
X_fft = X_fft[:, :n_coefs]
mcb = MultipleCoefficientBinning(n_bins=n_bins,
strategy=strategy,
alphabet=alphabet)
arr_desired = mcb.fit_transform(X_fft)
return arr_desired
import numpy as np
import matplotlib.pyplot as plt
from pyts.approximation import MultipleCoefficientBinning
# Parameters
n_samples, n_timestamps = 100, 12
# Toy dataset
rng = np.random.RandomState(41)
X = rng.randn(n_samples, n_timestamps)
# MCB transformation
n_bins = 3
mcb = MultipleCoefficientBinning(n_bins=n_bins, strategy='quantile')
X_mcb = mcb.fit_transform(X)
# Show the results for two time series
plt.figure(figsize=(6, 4))
plt.plot(X[0], 'o--', ms=4, label='First time series')
for x, y, s in zip(range(n_timestamps), X[0], X_mcb[0]):
plt.text(x, y, s, ha='center', va='bottom', fontsize=14, color='C0')
plt.plot(X[5], 'o--', ms=4, label='Second time series')
for x, y, s in zip(range(n_timestamps), X[5], X_mcb[5]):
plt.text(x, y, s, ha='center', va='bottom', fontsize=14, color='C1')
plt.hlines(mcb.bin_edges_.T, np.arange(n_timestamps) - 0.5,
np.arange(n_timestamps) + 0.5, color='g',
linestyles='--', linewidth=0.7)
def test_constant_sample():
"""Test that a ValueError is raised with a constant sample."""
discretizer = MultipleCoefficientBinning()
msg_error = "At least one timestamp is constant."
with pytest.raises(ValueError, match=msg_error):
discretizer.fit_transform(np.ones((10, 15)))