This is not a good solution. Those scorers provide an objective function
for grid search, and are standardised to report greater-is-better metrics.

On 4 February 2014 15:22, arahuja notifications@github.com wrote:

Setting greater_is_better=True causes the sign in _PredictScorer to be
negative. This leads to negative values when using LinearRegression in
cross_val_score

Alternatively, we could adjust the return statement in PredictScorer.
call_
return self._sign * self._score_func(y_true, y_pred, **self._kwargs)

Also, I found this mailthread on the issue:
https://www.mail-archive.com/scikit-learn-general@lists.sourceforge.net/msg08945.html

That mailthread implies that this is not the correct fix, but I wanted to

discuss how to correct this.

You can merge this Pull Request by running

git pull https://github.com/arahuja/scikit-learn regression-scorer-fix

Or view, comment on, or merge it at:

#2823
Commit Summary

• Fixing MAE and MSE errors. should not be negative when used in
  cross_val_score

File Changes

• M sklearn/metrics/scorer.pyhttps://github.com/Fixing MAE and MSE errors. should not be negative when used in cross_val #2823/files#diff-0(6)

Patch Links:

• https://github.com/scikit-learn/scikit-learn/pull/2823.patch
• https://github.com/scikit-learn/scikit-learn/pull/2823.diff

Reply to this email directly or view it on GitHubhttps://github.com//pull/2823
.

This is a known issue, see #2439.

But, indeed, your changes are not the right fix: they will result in grid search selecting the models with worst MSE.

So, closing this PR.