You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@madlib.apache.org by "ASF GitHub Bot (JIRA)" <ji...@apache.org> on 2017/12/06 20:15:00 UTC
[jira] [Commented] (MADLIB-1169) Change how cross validation stats
are reported and improve user docs
[ https://issues.apache.org/jira/browse/MADLIB-1169?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16280834#comment-16280834 ]
ASF GitHub Bot commented on MADLIB-1169:
----------------------------------------
GitHub user njayaram2 opened a pull request:
https://github.com/apache/madlib/pull/210
Regularized Regression: Change cross validation stats
JIRA:MADLIB-1169
Cross Validation seems to be supported by Elastic Net, SVM, and
Decision Trees. If a module is run with cross validation optimization
params, the output table corresponding to it displays `mean` and
`std` of the negative loss error for each permutation of the CV
params.
- This commit changes column names: `mean`->`mean_neg_loss` and
`std`->`std_neg_loss`.
- CV now uses negative Root Mean Squared Error, instead
of the negative Mean Squared Error.
- Update Elastic Net user docs to reflect these changes.
Closes #210
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/njayaram2/madlib improve/cross_validation_output
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/madlib/pull/210.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #210
----
commit ad41936c1a80e06cac6109690c15d7ad510c664d
Author: Nandish Jayaram <nj...@apache.org>
Date: 2017-12-06T19:58:46Z
Regularized Regression: Change cross validation stats
JIRA:MADLIB-1169
Cross Validation seems to be supported by Elastic Net, SVM, and
Decision Trees. If a module is run with cross validation optimization
params, the output table corresponding to it displays `mean` and
`std` of the negative loss error for each permutation of the CV
params.
- This commit changes column names: `mean`->`mean_neg_loss` and
`std`->`std_neg_loss`.
- CV now uses negative Root Mean Squared Error, instead
of the negative Mean Squared Error.
- Update Elastic Net user docs to reflect these changes.
Closes #210
----
> Change how cross validation stats are reported and improve user docs
> ---------------------------------------------------------------------
>
> Key: MADLIB-1169
> URL: https://issues.apache.org/jira/browse/MADLIB-1169
> Project: Apache MADlib
> Issue Type: Improvement
> Components: Module: Regularized Regression
> Reporter: Frank McQuillan
> Priority: Minor
> Fix For: v1.13
>
>
> Context
> Currently in cross validation, e.g.. for elastic net
> http://madlib.apache.org/docs/latest/group__grp__elasticnet.html
> CV stats are reported like:
> alpha | lambda_value | mean | std
> ------+--------------+---------------------+--------------------
> 0 | 100000 | -1.41777698585e+110 | 1.80536123195e+110
> 0.1 | 100000 | -1.19953054719e+107 | 1.72846143163e+107
> 1 | 100000 | -4175743937.91 | 2485189261.38
> etc.
> Here the "mean" col is the negative of the loss, which is a sort of accuracy, so the col header is not explanatory.
> Story
> As a MADlib developer, I want to make it clear what CV reported stats are, so that users are not confused as to what they mean.
> Acceptance
> 1) Change the calculation to report rmse from mse so that it is a smaller number that relates to the magnitude of the data. We can still report as negative, but just use rmse.
> 2) Rename the columns as "mean_neg_loss: and "std_neg_loss"
> 3) Improve the user docs to explain the col means for regression and classification.
> 4) Update any IC or functional/Tinc tests that are affected.
> 5) Update the example in
> http://madlib.apache.org/docs/latest/group__grp__elasticnet.html
> and also find any other modules where that need to be updated (SVM?)
> 6) Check the log_likelihood value reported in EN. Is it really a log likelihood and is it reported correctly?
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)