[
https://issues.apache.org/jira/browse/CLIMATE-481?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14050022#comment-14050022
]
Michael Joyce commented on CLIMATE-481:
---------------------------------------
Update, either of the stats functions that were used are fine, they both give
us the same thing. The one that we're currently using is probably a bit faster
since it doesn't return 5 different values, but I stuck with the one Kyo was
using for CLIMATE-483.
> StddevRatio and PatternCorr metrics outputting bad values
> ---------------------------------------------------------
>
> Key: CLIMATE-481
> URL: https://issues.apache.org/jira/browse/CLIMATE-481
> Project: Apache Open Climate Workbench
> Issue Type: Bug
> Components: metrics
> Affects Versions: 0.3-incubating
> Reporter: Michael Joyce
> Assignee: Michael Joyce
> Fix For: 0.4
>
>
> Kyo reported the StdDevRatio and PatternCorr metrics were returning
> unexpected results. He provided some example code (below) that demonstrated
> expected values.
> There seems to be a few problems with the current metrics. First, we don't
> have a version of these that doesn't do some sort of rebin on the data first.
> That should be changed (in fact, we should probably only have versions that
> don't rebin the data, but that's a separate conversation perhaps). Second,
> StdDevRatio currently normalizes to the target, instead of the reference
> dataset. That's an easy fix. Third, Kyo is using a different stats method to
> get what he wants. I will play around a bit and see if I can determine which
> we should use and get it changed.
> {code}
> taylor_data[imodel, 0]=ma.std(model.values)/ma.std(ref_dataset.values)
> taylor_data[imodel, 1]=stats.mstats.linregress(ref_dataset.values,
> model.values)[2]
> {code}
--
This message was sent by Atlassian JIRA
(v6.2#6252)
