simplify grab_residuals

[dajmcdon]

closes #270 .

[dsweber2]

oh, it returns NULL if there aren't residuals? that does make this way easier. I'm assuming there's at least one non-regression model calling grab_residuals somewhere in the tests?

[dajmcdon]

I kind of doubt that there is a test where a classifier calls this function. Are you thinking that there needs to be?

[dshemetov]

Can we bump minor/patch version and add a note to CHANGELOG?

[dajmcdon]

@dshemetov Sure, though I'm unfamiliar. What would you like to bump to? 0.0.7 or 0.0.6001? Something else?

[dshemetov]

0.0.7 sounds good to me.

[dshemetov]

This might do it for you (haven't tried).

[dsweber2]

er, I was actually thinking was a model for which stats:residuals isn't defined (is it for quant_reg, for example?). Totally wrote the wrong thing sorry.

[dajmcdon]

residuals are just the difference between the training response and the point prediction in-sample (at least the kind we're using here), so would be mathematically defined for any regression engine (including quantile regression). This should fail if the engine is a classifier though.