Bug fix - test_csvs argument was ignored

[Jendker]

Removed spurious overwriting of argument 'test_csvs' in evaluate.py
This bug leads to problems in lm_optimizer.py

[community-tc-integration]

No Taskcluster jobs started for this pull request

The `allowPullRequests` configuration for this repository (in `.taskcluster.yml` on the
default branch) does not allow starting tasks for this pull request.

[kdavis-mozilla]

It's used on the very next line.

[Jendker]

The variable is used on the next line, that's fine.

But the function argument has the same variable name and is overwritten.
This leads to problems in lm_optimizer.py where each time all test_csvs are used, because given argument is ignored.

[kdavis-mozilla]

Have you checked all other calls to evaluate() to make sure they don't rely on this overwriting?

[Jendker]

Yes, I've checked it. Everywhere else as argument FLAGS.test_files.split(',') was given, so the bug was hidden. Other case is only in lm_optimizer.py, where this fix corrects the pruner behaviour.

[kdavis-mozilla]

Hmm.

Looks like the only other calls are train.py:643, evaluate.py:143, and lm_optimizer.py:36. All of those look like they don't rely on the overwriting and the only one which breaks as a result of the overwriting is lm_optimizer.py:36.

So you're change looks good.

[kdavis-mozilla]

LGTM

[kdavis-mozilla]

Thanks!