Enable label alignment for token classification datasets

[lewtun]

This PR extends the Dataset.align_labels_with_mapping() method to support alignment of label mappings between datasets and models for token classification (e.g. NER).

Example of usage:

from datasets import load_dataset

ner_ds = load_dataset("conll2003", split="train")
# returns [3, 0, 7, 0, 0, 0, 7, 0, 0]
ner_ds[0]["ner_tags"]
# hypothetical model mapping with O <--> B-LOC
label2id = {
    "B-LOC": "0",
    "B-MISC": "7",
    "B-ORG": "3",
    "B-PER": "1",
    "I-LOC": "6",
    "I-MISC": "8",
    "I-ORG": "4",
    "I-PER": "2",
    "O": "5"
  }
ner_aligned_ds = ner_ds.align_labels_with_mapping(label2id, "ner_tags")
# returns [3, 5, 7, 5, 5, 5, 7, 5, 5]
ner_aligned_ds[0]["ner_tags"]

Context: we need this in AutoTrain to automatically align datasets / models during evaluation. cc @abhishekkrthakur

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[mariosasko]

Two nits.

PS: It makes sense to also add support for the Sequence type to class_encode_column. I can work on that.

[lewtun]

Hmm, not sure why the Windows tests are failing with:

Did not find path entry C:\tools\miniconda3\bin
C:\tools\miniconda3\envs\py37\python.exe: No module named pytest

Edit: running the CI again fixed the problem 🙃

[mariosasko]

One last nit and we can merge then

[lewtun]

One last nit and we can merge then

Thanks, done!

[mariosasko]

Looks good now. Thanks!