pydtw

pydtw is a simple python wrapper of libdtw, which is a fast, dynamic time warping library based on the UCR Suite.

Works for Python 3, but not for Windows, because I don't understand the Windows C compiler.

Install

pip install -e .

Usage

import numpy as np
import dtw


data = np.cumsum(np.random.uniform(-0.5, 0.5, 1000000))
query = np.cumsum(np.random.uniform(-0.5, 0.5, 100))
results = dtw.query(data, query, r=0.05)

• input
  • data : numpy.array of data
  • query : numpy.array of the pattern to be matched
  • r : size of warping window
• result : dict
  • results["index"] : the index of the first element in the best matching sequence in the data.
  • results["value"] : the DTW distance between the query and the matching sequence in the data.

Known Issues

• The first element of the time-series is removed when translating from Python to C. I dont't know enough about Python -> C to fix this and it doesn't really matter for my purposes.