[enhancement] refactor bert finetuning script

[eric-haibin-lin]

Description

summary of changes:

• introduce GlueTask class which stores the metadata of the task (class labels, is_pair, etc)
• reuse the Dataset API for glue datasets introduced in [Dataset] Fix CoLA dataset index and add Glue dataset API  #682
• fix bert transform for the case when label is not available
• add script test

Checklist

Essentials

• PR's title starts with a category (e.g. [BUGFIX], [MODEL], [TUTORIAL], [FEATURE], [DOC], etc)
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage
• Code is well-documented

Changes

• Feature1, tests, (and when applicable, API doc)
• Feature2, tests, (and when applicable, API doc)

Comments

• If this change is a backward incompatible change, why must this change be made.
• Interesting edge cases to note here

[eric-haibin-lin]

@Gpwner here is the fix for the BERTTransform bug. Thanks for reporting

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@b079bab). Click here to learn what that means.
The diff coverage is 100%.

@@           Coverage Diff            @@
##             master    #692   +/-   ##
========================================
  Coverage          ?   89.4%           
========================================
  Files             ?      66           
  Lines             ?    5918           
  Branches          ?       0           
========================================
  Hits              ?    5291           
  Misses            ?     627           
  Partials          ?       0

Flag	Coverage Δ
#PR692	89.4% <100%> (?)
#notserial	64.65% <50%> (?)
#py2	89.4% <100%> (?)
#serial	68.06% <100%> (?)

[codecov]

Codecov Report

❗ No coverage uploaded for pull request base (master@b079bab). Click here to learn what that means.
The diff coverage is 100%.

@@            Coverage Diff            @@
##             master     #692   +/-   ##
=========================================
  Coverage          ?   90.94%           
=========================================
  Files             ?       64           
  Lines             ?     5887           
  Branches          ?        0           
=========================================
  Hits              ?     5354           
  Misses            ?      533           
  Partials          ?        0

[tlby]

Good stuff, looks worthwhile! I've hacked in a couple of these features myself downstream (mp for the BERTDatasetTransform and GPU selection).

[vanewu]

LGTM 👍

[mli]

Job PR-692/10 is complete.
Docs are uploaded to http://gluon-nlp-staging.s3-accelerate.dualstack.amazonaws.com/PR-692/10/index.html

[pengxin99]

I try the new script, and found the error:
@eric-haibin-lin

INFO:root:processing dataset...
multiprocessing.pool.RemoteTraceback:
"""
Traceback (most recent call last):
  File "/home/pengxiny/anaconda3/envs/mxnet-cpu/lib/python3.6/multiprocessing/pool.py", line 119, in worker
    result = (True, func(*args, **kwds))
  File "/home/pengxiny/anaconda3/envs/mxnet-cpu/lib/python3.6/multiprocessing/pool.py", line 44, in mapstar
    return list(map(*args))
  File "/home/pengxiny/autotest_gluonnlp/gluon-nlp/scripts/bert/dataset.py", line 432, in __call__
    return self._bert_xform(line)
  File "/home/pengxiny/anaconda3/envs/mxnet-cpu/lib/python3.6/site-packages/gluonnlp-0.6.0-py3.6.egg/gluonnlp/data/transforms.py", line 1089, in __call__
    assert len(line) == 2
AssertionError
"""

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "finetune_classifier.py", line 328, in <module>
    bert_tokenizer, task, batch_size, dev_batch_size, args.max_len, args.pad)
  File "finetune_classifier.py", line 314, in preprocess_data
    data_test = mx.gluon.data.SimpleDataset(pool.map(test_trans, data))
  File "/home/pengxiny/anaconda3/envs/mxnet-cpu/lib/python3.6/multiprocessing/pool.py", line 266, in map
    return self._map_async(func, iterable, mapstar, chunksize).get()
  File "/home/pengxiny/anaconda3/envs/mxnet-cpu/lib/python3.6/multiprocessing/pool.py", line 644, in get
    raise self._value
AssertionError

[szha]

@pengxin99 you may need to install GluonNLP from master branch first. http://gluon-nlp.mxnet.io/install.html#install-from-github

[eric-haibin-lin]

@pengxin99 would you mind sending me the complete command to reproduce it?

[pengxin99]

@szha Thanks and i think i run the script at the latest gluonnlp(use python setup.py install)

@eric-haibin-lin I list the env and run command below:
OS: CentOS Linux release 7.4.1708 (Core)
mxnet: pip install mxnet-mkl --pre
gluonnlp: python setup install (latest version)

finetune: GLUE_DIR=glue_data python finetune_classifier.py --task_name MRPC --batch_size 32 --optimizer bertadam --epochs 3 --lr 2e-5

[pengxin99]

I use the #708 to have a try, and it works well :)
thanks, @eric-haibin-lin @szha