Dynamic Time Warping Python Module | 中文
Dynamic time warping is used as a similarity measured between temporal sequences. This package provides two implementations:
- the basic version (see here) for the algorithm
- an accelerated version which relies on scipy cdist (see pollen-robotics#8 for detail)
import numpy as np
# We define two sequences x, y as numpy array
# where y is actually a sub-sequence from x
x = np.array([2, 0, 1, 1, 2, 4, 2, 1, 2, 0]).reshape(-1, 1)
y = np.array([1, 1, 2, 4, 2, 1, 2, 0]).reshape(-1, 1)
from dtw import dtw
euclidean_norm = lambda x, y: np.abs(x - y)
d, cost_matrix, acc_cost_matrix, path = dtw(x, y, dist=euclidean_norm)
print(d)
>>> 2.0 # Only the cost for the insertions is kept
# You can also visualise the accumulated cost and the shortest path
import matplotlib.pyplot as plt
plt.imshow(acc_cost_matrix.T, origin='lower', cmap='gray', interpolation='nearest')
plt.plot(path[0], path[1], 'w')
plt.show()
Result of the accumulated cost matrix and the shortest path (in white) found:
python -m pip install dtw
It is tested on Python 2.7, 3.4, 3.5 and 3.6. It requires numpy and scipy.