hfawaz / aaltd18

Data augmentation using synthetic data for time series classification with deep residual networks
GNU General Public License v3.0
183 stars 42 forks source link

Definition of warping window [DTW_PARAMS] #3

Closed petteriTeikari closed 5 years ago

petteriTeikari commented 5 years ago

Hi again @hfawaz et al.

I was wondering if you used actually the following setting for your paper, i.e. you did not warp at all if I understood the code correctly and read the paper properly?

DTW_PARAMS = {'w':-1} warping window should be given in percentage (negative means no warping window)

In the paper [6, Forestier et al. 2017, PDF] cited, that was given as following for the "old-school" Windows Warping (WW) method

The Windows Warping (WW) method discussed in [11] involves warping a randomly selected slice of a time series by speeding it up or down. We follow the recommendations given in [11] and use warping ratios equal to either 1/2 or 2 on slices representing 10% of time series’s length

image

Did you test out the warping with DTW?

hfawaz commented 5 years ago

Hello @petteriTeikari,

Thank you for your interest, I am sorry if the comment in the code was misleading.

By no "warping window", we meant that the warping could be infinite (there was no boundaries).

So to answer the question: we did allow warping - this is the only setting we have tested.

I hope this answers your inquiry.

petteriTeikari commented 5 years ago

I see, thanks for the clarification @hfawaz