This example illustrates the principle of time series envelope and its relationship to the “LB_Keogh” lower bound [1].
The envelope of a time series consists of two time series such that the original time series is between the two time series. Denoting the original time series \(X = (X_i)_{1 \leq i \leq n}\), the envelope of this time series is an ensemble of two time series of same length \(L = (l_i)_{1 \leq i \leq n}\) and \(U = (u_i)_{1 \leq i \leq n}\) such that for all \(i \in \{1, \ldots, n\}\):
\[ \begin{align}\begin{aligned}u_i = \max(x_{i - r}, \ldots, x_{i + r})\\l_i = \min(x_{i - r}, \ldots, x_{i + r})\end{aligned}\end{align} \]
where \(r\) is the radius of the envelope.
The distance between a time series $Q$ and an envelope \((L, U)\) is defined as:
\[\begin{split}LB_{Keogh}(Q, (L, U)) = \sqrt{\sum_{i=1}^n \begin{cases} (q_i - u_i)^2 & \text{if $q_i > u_i$}\\ (q_i - l_i)^2 & \text{if $q_i < l_i$}\\ 0 & \text{otherwise} \end{cases} }\end{split}\]
So it is simply the Euclidean distance between \(Q\) and the envelope.
[1] E. Keogh and C. A. Ratanamahatana, “Exact indexing of dynamic time warping”. Knowledge and Information Systems, 7(3), 358-386 (2004).
# Author: Romain Tavenard
# Johann Faouzi
# License: BSD 3 clause
# sphinx_gallery_thumbnail_number = 2
import numpy
import matplotlib.pyplot as plt
from tslearn.generators import random_walks
from tslearn.preprocessing import TimeSeriesScalerMeanVariance
from tslearn import metrics
numpy.random.seed(0)
n_ts, sz, d = 2, 100, 1
dataset = random_walks(n_ts=n_ts, sz=sz, d=d)
scaler = TimeSeriesScalerMeanVariance(mu=0., std=1.) # Rescale time series
dataset_scaled = scaler.fit_transform(dataset)
plt.figure(figsize=(14, 8))
envelope_down, envelope_up = metrics.lb_envelope(dataset_scaled[0], radius=3)
plt.plot(dataset_scaled[0, :, 0], "r-", label='First time series')
plt.plot(envelope_down[:, 0], "b-", label='Lower envelope')
plt.plot(envelope_up[:, 0], "g-", label='Upper envelope')
plt.legend()
plt.title('Envelope around a time series with radius=3')
plt.figure(figsize=(14, 8))
plt.plot(envelope_down[:, 0], "b-", label='Lower envelope')
plt.plot(envelope_up[:, 0], "g-", label='Upper envelope')
plt.plot(dataset_scaled[1, :, 0], "k-", label='Second time series')
plt.vlines(numpy.arange(sz), dataset_scaled[1, :, 0], numpy.clip(
dataset_scaled[1, :, 0], envelope_down[:, 0], envelope_up[:, 0]),
label='Distance', color='orange')
plt.legend()
lb_k_sim = metrics.lb_keogh(dataset_scaled[1],
envelope_candidate=(envelope_down, envelope_up))
plt.title('Distance between the second time series and \n'
'the envelope = {:.4f}'.format(lb_k_sim))
plt.show()
Total running time of the script: (0 minutes 0.785 seconds)
Gallery generated by Sphinx-Gallery
RetroSearch is an open source project built by @garambo | Open a GitHub Issue
Search and Browse the WWW like it's 1997 | Search results from DuckDuckGo
HTML:
3.2
| Encoding:
UTF-8
| Version:
0.7.4