If its ok with everyone I'll come up with a proposal to float in the dev list
and create a JIRA item. If there are any thoughts or discussions around the
runtime performance module that you guys have already had I'd be much obliged
if you let me know.
My thoughts around the runtime perf module:
In a nutshell the module could: 1) take in an algorithm (or a set of
algorithms) and a combination of features 2) measure latency, run time
performance, cpu load, io characteristics of the algorithm 3) spit the result
out to some visualization dashboard (can we use zeppelin for this). I imagine
it would be something that mahout committers could run every release before and
after any new code gets submitted. One other point to consider is whether the
performance will be solely for samsara or a dashboard of performance metrics
for samsara connected to each of the backends.
Thoughts?
Proposal to come in the next week or so.
Thanks in advance.
> From: [email protected]
> To: [email protected]
> Subject: RE: Mahout contributions for next sprint
> Date: Fri, 13 May 2016 21:41:57 +0000
>
> Hello Saikat
> We actually have precision and recall metrics built into our confusion matrix:
>
> https://github.com/apache/mahout/blob/master/math-scala/src/main/scala/org/apache/mahout/classifier/stats/ConfusionMatrix.scala
>
> Embarrassed to say that I forgot that myself.
>
> The runtime performance module does sound like it would be very useful.
>
> Andy
>
>
>
> -------- Original message --------
> From: Saikat Kanjilal <[email protected]>
> Date: 05/13/2016 5:10 PM (GMT-05:00)
> To: [email protected]
> Subject: Mahout contributions for next sprint
>
> Hello Mahout Committers,I spoke to AndrewM and he suggested that I work on
> some precision/recall or run-time performance module, or possibly streaming
> kmeans (apparently there's a new JIRA associated with this). I was wondering
> which of these would be the most impactful and interesting to the project.
> I would be glad to help in any of these areas.Thanks in advance.
