Option to specify text col name in TextClassificationProcessor and RegressionProcessor

[PhilipMay]

PR for #379

TODO

• check if and how tests can be written -> added parametrize to the tests

[Timoeller]

Love it. Having tests also helps a lot. Thanks for this great addition.

I added a small comment in the column renaming.
Lets wait for our CI issue to be fixed, than we can merge it safely.

[Timoeller]

Could you do a dummy commit to restart the CI, please? (I do not have access to the pipeline, today is public holiday in Germany...)

[PhilipMay]

rebased master

[PhilipMay]

Fixed test_processor_saving_loading the same way as #390 does.

[PhilipMay]

The CI system was gracious to me and all tests are green now. :-)

Can you please do a review?

Thanks
Philip

[Timoeller]

One small check is needed, than we can merge.

I would also merge irrespective of our CI status, since the pipeline ran through already. The problems seem to come from Azure side. Apparently there are not enough resources for open source code at all times and we currently investigate alternatives like github actions.

[PhilipMay]

Made the change discussed above.

I am passing usecols=columns_needed to pd.read_csv to read only those cols we need. Right from the beginning. This way the line df = df[columns_needed] is not needed anymore.

The advantage is memory usage and early crashing. If one of the cols does not exist (because the user mistyped it) pd.read_csv crashes with a meaningful error. Example: ValueError: Usecols do not match columns, columns expected but not found: ['text_other_']

I think this should adress the concerns from @Timoeller in the review.

[PhilipMay]

rebased master (mainly to trigger CI again)

[PhilipMay]

YAY! CI is green. :-)

[Timoeller]

CSV data reading is handled much more user friendly now.
At some point we need to adjust the read_tsv_sentence_pair() to reflect these changes as well.

Thanks @PhilipMay for your effort.