Test coverage of special parameterizations

[sdmccabe]

There are some methods that have parameterizations that have different behavior; these are not always tested explicitly in the test suites, where we tend to favor the defaults. Examples:

• Correlation matrix (regularization)
• Laplacian distances (kernel, norm, etc.)

I'm sure there are others. Ideally we would have a full account of these and make sure they are fully covered by the tests.

[leotrs]

You mean execute the same tests as we have now over different parameterizations? Or do you mean having specialized tests that cover a range of parameters?

[sdmccabe]

If specialized tests are warranted, then that's fine, but I really just mean executing the same tests to make sure the parameterizations work at all.

[leotrs]

We could have the tests iterate over a manually-built list of methods that includes different parameterizations of the same class.

…

On Wed, May 22, 2019 at 12:00 PM Stefan McCabe ***@***.***> wrote: If specialized tests are warranted, then that's fine, but I really just mean executing the same tests to make sure the parameterizations work at all. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#202?email_source=notifications&email_token=AAILYAHFVKECODG6JCOF4ZDPWVU3TA5CNFSM4HOVTAK2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODV7Q6JI#issuecomment-494866213>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAILYAGXSGPRYKCMFNH3VULPWVU3TANCNFSM4HOVTAKQ> .

-- www.leotrs.com | [email protected] PhD student at the Network Science Institute, Northeastern University