search

asardaes / dtwclust

R Package for Time Series Clustering Along with Optimizations for DTW
https://cran.r-project.org/package=dtwclust
GNU General Public License v3.0
254 stars 29 forks source link

Does dtw_basic do open-ended alignments by default? #51

Closed masonyoungblood closed 2 years ago

masonyoungblood commented 3 years ago

I'm using dtw_basic with proxy::dist, and after hierarchical clustering (with a variety of different parameter combinations) I'm getting clusters where very short signals are getting matched with very long signals that have a similar sub-sequence. It seems as though dtw_basic might be doing open-ended alignment. Is that the case? I can't find anything about it in the documentation.

I'm using a window size of 5-10% (of the average signal length), so I'm not sure how else these signals would be clustered together. Let me know if you have any thoughts!

asardaes commented 3 years ago

Hello, no, the dtw_basic version doesn't do open-end alignments. Are you setting normalize = TRUE?

masonyoungblood commented 3 years ago

Okay thanks, and yeah I'm using normalize - Could that be inflating the similarity between signals of different lengths?

On Sun, Nov 29, 2020, 4:37 PM Alexis Sardá notifications@github.com wrote:

Hello, no, the dtw_basic version doesn't do open-end alignments. Are you setting normalize = TRUE?

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/asardaes/dtwclust/issues/51#issuecomment-735461814, or unsubscribe https://github.com/notifications/unsubscribe-auth/AKOFD2R2NZDTZZNTBBE3XTDSSK5KRANCNFSM4UGXH7EQ .

asardaes commented 3 years ago

No, it's the other way around, that's why I wanted to know. It could also be that the distance between those long/short series is high, but there's just no other centroid with more similarity.