-
Notifications
You must be signed in to change notification settings - Fork 67
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
getEuclideanDistance should depends on the state of timeseries #28
Comments
Thank you for pointing this out. There are two cases:
Typically, we force the query to be z-normed in this case. However, I agree that it is possible to leave out z-norm, too. The proposed pull-request however, breaks the code by removing z-norm for subsequences. |
For my understanding, it's better to return the un-normalized distance measure. So a potential fix will be: if the query is normalized -> unnormalize its values, if the timeseries is normalized -> un-normalize the values -> then calculate the euclidean distance between original un-normalized versions of both timeseries data and query :) |
Removing it, is not sooo easy. For example, SFA uses the momentary Fourier transform. The MFT applies z-normalization to each subsequence: SFA/src/main/java/sfa/transformation/MFT.java Line 112 in 43cfad8
The same does So, when removing it from |
yes i know, that's why i didn't propose a fix. I saw that in subsequence matching, each subsequence is normalized with different mean and std, so i can't un-normalized. maybe we need 2 separate methods. |
The getEuclidean Distance gives incorrect results.
It should depends whether the timeseries is normalized or not and whether the query is normalized or not.
In case of normalization, both should be de-normalized (value * std + avg).
The current implementation normalize again !
As well, for the epsilon radius search, std needs to be taken into account
Cheers!
Assaad
The text was updated successfully, but these errors were encountered: