[R-package] fix R examples and lgb.plot.interpretation()

[jameslamb]

See comments from @StrikerRUS

• [R-package] [docs] Simplified examples to cut example run time (fixes #2988) #2989 (comment)
• [R-package] [docs] Simplified examples to cut example run time (fixes #2988) #2989 (comment)

A few of the R examples are broken right now. These weren't caught by tests because one is only a warning, one is a plot that is generated successfully but is incorrect, and one is an issue specific to our readthedocs builds.

I think the issue for lgb.cv() is #2715. In that issue we found that data.table 1.11.4 and R 3.6.0 lead to a data.table error ending in column or argument 2 is NULL. I think this can be fixed by upgrading to the data.table 1.12.1 in conf.py.

To fix the lgb.set.categorical() example, I changed file names across several examples to be sure that no two examples write to the same file name.

This issue with lgb.plot.interpretation() is because I made a change and forgot to use abs().

new plot:

I was not able to reproduce the data.table error in #2989 (comment). It

[StrikerRUS]

Thanks @jameslamb for taking a look!

I think the issue for lgb.cv() is #2715.

But we do not use lgb.dl() at RTD and R version is 3.5 there. Probably we should update our FAQ entry...

If you work from LightGBM's branch, not from your fork, then I'll be able to enable RTD builds for that branch and make sure that all issues are gone. Will it make sense?

[jameslamb]

Thanks @jameslamb for taking a look!

I think the issue for lgb.cv() is #2715.

But we do not use lgb.dl() at RTD and R version is 3.5 there. Probably we should update our FAQ entry...

If you work from LightGBM's branch, not from your fork, then I'll be able to enable RTD builds for that branch and make sure that all issues are gone. Will it make sense?

Sure! I can recreate the PR from a LightGBM branch.

[jameslamb]

closing in favor of #3002