mahout-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Saikat Kanjilal <sxk1...@hotmail.com>
Subject mahout-1869--[discussion on design of next steps]
Date Sat, 24 Sep 2016 18:46:57 GMT
Dmitry/AndrewM,

I'm about to move forward finally with the next steps for this, to start with I was just going
to pick running performance numbers around the classifiers in math-scala module, specifically
naivebayes, I was also going to add the perf hook around tfidf.    However I was thinking
it might be more useful to be able to shim in instrumentation around each parts of the above
algorithm to measure the performance.   The goal of this component is to measure the performance
of not only the whole algorithm but each significant part of the algorithm.   A more concrete
use case might be that the comitter makes a change to several different parts of an algorithm
and wants to do a before and after analysis.



Does the above path sound like a suitable first stab,  I have the basic perf stats printing
out and am stepping back a bit to think about the overall design and applicability into the
various scala/samsara parts of mahout.


I would really like some input before I move in a particular direction.


Thanks in advance.

Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message