[R-package] factored dependency 'magrittr' out of R package

[jameslamb]

@Laurae2 in this PR, I propose that we completely factor out the dependency magrittr, in the interest of reducing lightgbm's dependency footprint. I know magrittr itself has minimal additional dependencies, but in general I think we should strive to keep lightgbm as light as possible and I feel this particular dependency isn't providing a lot of value.

[jameslamb]

Just rebased this to master to keep up with recent changes.

[StrikerRUS]

@jameslamb Can we merge this? If so, I'll be able to remove magrittr dependency from #2176, which is going to be merged this weekend.

[jameslamb]

@jameslamb Can we merge this? If so, I'll be able to remove magrittr dependency from #2176, which is going to be merged this weekend.

Seems that one file is in conflict. Let me rebase and run the tests manually (because I haven't done #2335 yet).

[jameslamb]

Ok we currently have 4 failing tests:

══ testthat results  ═══════════════════════════════════════════════════════════
[ OK: 33 | SKIPPED: 0 | WARNINGS: 3 | FAILED: 4 ]
1. Error: training continuation works (@test_basic.R#71)
2. Failure: Feature penalties work properly (@test_parameters.R#36)
3. Failure: Feature penalties work properly (@test_parameters.R#37)
4. Failure: Feature penalties work properly (@test_parameters.R#38)

The "training continuation" one exists on master, but I think the other three are problems I've introduced in this PR. I don't feel comfortable merging until I address those three. WIll try to do it today.

[jameslamb]

WOW I think this is why the tests were failing for me on this build. typo: split_feaure 😆

testing now

[jameslamb]

Ok we currently have 4 failing tests:

══ testthat results  ═══════════════════════════════════════════════════════════
[ OK: 33 | SKIPPED: 0 | WARNINGS: 3 | FAILED: 4 ]
1. Error: training continuation works (@test_basic.R#71)
2. Failure: Feature penalties work properly (@test_parameters.R#36)
3. Failure: Feature penalties work properly (@test_parameters.R#37)
4. Failure: Feature penalties work properly (@test_parameters.R#38)

The "training continuation" one exists on master, but I think the other three are problems I've introduced in this PR. I don't feel comfortable merging until I address those three. WIll try to do it today.

Alright I've figured out the three test errors I introduced. R unit tests now all work. I need to fix the "training continuation works" one but this PR doesn't need to wait on that.

@Laurae2 could you take one more look at this PR and let me know if I could merge?

[StrikerRUS]

@jameslamb Please remove this dependency from here as well.

LightGBM/docs/conf.py

Line 232 in e776582

r-magrittr=1.5=r351h6115d3f_4 \

[jameslamb]

@jameslamb Please remove this dependency from here as well.

LightGBM/docs/conf.py

Line 232 in e776582

r-magrittr=1.5=r351h6115d3f_4 \

done. Good catch, I forgot we do that. I did a git grep magrittr just to be sure there are no other uses in non-standard places, did not find any.

[StrikerRUS]

@jameslamb

done. Good catch, I forgot we do that. I did a git grep magrittr just to be sure there are no other uses in non-standard places, did not find any.

Thanks!

[jameslamb]

I just rebased this PR to master.

Tests for the R package are passing locally (see below)

export CXX=/usr/local/bin/g++-8 CC=/usr/local/bin/gcc-8
Rscript build_r.R
cd R-package/tests
Rscript testthat.R

Once all CI checks pass, I'm going to merge this.