Enforce stricter NLTK version.

[mrdrozdov]

Certain tagging apis break in more recent versions of NLTK.

[japerk]

I've tested this with NLTK 3 before. Can you tell me what version broke & how?

[mrdrozdov]

Was using the most recent version of NLTK. Broke when running this command.

python train_tagger.py treebank --brill

$ python --version
Python 2.7.9
$ pip --version
pip 7.1.2 from /Users/Andrew/Developer/nltk-trainer/.test-env/lib/python2.7/site-packages (python 2.7)
$ virtualenv --version
12.1.1
$ pip install -r requirements.txt
Collecting argparse>=1.1 (from -r requirements.txt (line 1))
  Using cached argparse-1.4.0-py2.py3-none-any.whl
Collecting nltk>=2.0b8 (from -r requirements.txt (line 2))
  Using cached nltk-3.1.tar.gz
Collecting numpy>=1.3.0 (from -r requirements.txt (line 3))
  Using cached numpy-1.10.1-cp27-none-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl
Collecting scipy>=0.7.0 (from -r requirements.txt (line 4))
  Using cached scipy-0.16.1-cp27-none-macosx_10_6_intel.macosx_10_9_intel.macosx_10_9_x86_64.macosx_10_10_intel.macosx_10_10_x86_64.whl
Installing collected packages: argparse, nltk, numpy, scipy
  Running setup.py install for nltk
Successfully installed argparse-1.4.0 nltk-3.1 numpy-1.10.1 scipy-0.16.1
$ pip freeze
docopt==0.6.1
nltk==3.1
numpy==1.10.1
scipy==0.16.1
virtualenv==12.0.7
$ python train_tagger.py treebank --brill
loading treebank
3914 tagged sents, training on 3914
training AffixTagger with affix -3 and backoff <DefaultTagger: tag=-None->
training <class 'nltk.tag.sequential.UnigramTagger'> tagger with backoff <AffixTagger: size=2536>
training <class 'nltk.tag.sequential.BigramTagger'> tagger with backoff <UnigramTagger: size=4940>
training <class 'nltk.tag.sequential.TrigramTagger'> tagger with backoff <BigramTagger: size=2328>
Traceback (most recent call last):
  File "train_tagger.py", line 264, in <module>
    trace=args.trace, max_rules=args.max_rules, min_score=args.min_score)
  File "/Users/Andrew/Developer/nltk-trainer/nltk_trainer/tagging/training.py", line 7, in train_brill_tagger
    brill.SymmetricProximateTokensTemplate(brill.ProximateTagsRule, *bounds),
AttributeError: 'module' object has no attribute 'SymmetricProximateTokensTemplate'

[mrdrozdov]

Have read elsewhere that Brill Tagger (maybe all taggers?) have changed API in NLTK 3.

[japerk]

Ok, thanks for the details. I think everything else should work with NLTK 3, but I'll work on an update for brill.