henrikingo opened a new issue, #152:
URL: https://github.com/apache/otava/issues/152

   We have a lot of activity from new contributors now, and some patches have 
either direct dependencies, or nevertheless makes sense to implement in a 
certain order. So I'm creating this ticket purely to be used as a parent ticket 
to those sub tasks, and to centralize any discussion across tickets that might 
be necessary. I will summarize the main point of each sub-task here:
   
   
   
   #150 adds an option to define the minimum number of points that have to 
exist before a change point can be found.
    - This actually exists in "by the book" e-divisive and is necessary for 
some proof of convergence
    - MongoDB set this to a hard-coded 2 points. (In otjher words, minimum 
length of a window or sequence was 5)
    - The reason why this has been omitted in the Hunter version of the code 
base, is that this would prevent correctly finding both change points, if there 
are two changes closely after each other,.
    - In my opinion doing this on top of the split-merge version of the alorith 
doesn't make sense, since is the whole point with dividing the time into many 
small chunks series int                                                         
                                                                                
                                                                                
                                                                                
                                                
   
   (coming soon) Add a new "composition" of the e-divisive, from various 
features that already exist.
   - Short version: Like vanilla e-divisive, but using Student t-test in the 
end.
   - Based on hypothesis that the split-merge approach was largely needed and 
addressing the bug that was found and fixed by Sowiks and is discussed in #96 
   
   #141 fixes a bug in master that where the root cause is that we have 2 
different ChangePoint classes now and with --orig-edivisive at some point they 
get mixed up and eventually crash the app. 
   - The previous ticket in this list also hits this bug.
   
   #151 I created this to highlight and understand the root cause behind #141. 
   - In the meantime, I think #141 is a good "workaround'ish" way to solve it. 
In fact I'd argue it nicely highlights where the root cause fix should look 
into thiis
   - 


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to