Need help

[Okroshiashvili]

• Python port of R's Comprehensive Dynamic Time Warp algorithm package version: 1.1.6
• Python version: 3.7
• Operating System: Ubuntu 18.04

Description

I want to find DTW score between two time series. To understand how these algorithm work I use two libraries, dtw-python and tslearn.

However, I'm getting different results and I don't know why. Any help would be appreciate :)

What I Did

import numpy as np
import tslearn.metrics as tsm
import dtw

x = np.linspace(0, 50, 100)

ts1 = pd.Series(3 * np.sin(x / .5))
ts2 = pd.Series(2 * np.sin(x))

Results

tsm.dtw(ts1, ts2) -> 16.578554103357583

dtw_result = dtw.dtw(x=ts1, y=ts2, dist_method="euclidean", step_pattern="symmetric2") -> 174.2896

Why do I have such a big difference? I tried all the available step_pattern functions for dtw-python library but without any success. Which one is correct result?

[tonigi]

Try with a small alignment that can be computed by hand. In the documentation and tests there should be at least one. Il mar 16 feb 2021, 15:21 Nodar Okroshiashvili <notifications@github.com> ha scritto:

…

- Python port of R's Comprehensive Dynamic Time Warp algorithm package version: 1.1.6 - Python version: 3.7 - Operating System: Ubuntu 18.04 Description I want to find DTW score between two time series. To understand how these algorithm work I use two libraries, dtw-python and tslearn <https://github.com/tslearn-team/tslearn>. However, I'm getting different results and I don't know why. Any help would be appreciate :) What I Did import numpy as np import tslearn.metrics as tsm import dtw x = np.linspace(0, 50, 100) ts1 = pd.Series(3 * np.sin(x / .5)) ts2 = pd.Series(2 * np.sin(x)) Results tsm.dtw(ts1, ts2) -> 16.578554103357583 dtw_result = dtw.dtw(x=ts1, y=ts2, dist_method="euclidean", step_pattern="symmetric2") -> 174.2896 Why do I have such a big difference? I tried all the available step_pattern functions for dtw-python library but without any success. Which one is correct result? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#9>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAU74C2OAUPPCVAMTAQE4ATS7J5PDANCNFSM4XWRDH5Q> .

[Okroshiashvili]

Sorry @tonigi I did not understand what you mean. Can you elaborate little bit more?

[tonigi]

. The algorithm can be "run" with pen and paper on small cases. I was suggesting to check which implementation is correct (maybe both, with different conventions) using a "manual" solution as a reference.

I can't debug other people's implementations.