[recipe request] AllenNLP

[jayfurmanek]

• Name of package: AllenNLP
• URL of package source code: https://github.com/allenai/allennlp
• License of package: Apache-2
• List of package dependencies:

torch>=0.4.1,<1.2

# Parameter parsing (but not on Windows).
jsonnet>=0.10.0 ; sys.platform != 'win32'

# Adds an @overrides decorator for better documentation and error checking when using subclasses.
overrides

# Used by some old code.  We moved away from it because it's too slow, but some old code still
# imports this.
nltk

# Mainly used for the faster tokenizer.
spacy>=2.1.0,<2.2

# Used by span prediction models.
numpy

# Used for reading configuration info out of numpy-style docstrings.
numpydoc>=0.8.0

# Used in coreference resolution evaluation metrics.
scipy
scikit-learn

# Write logs for training visualisation with the Tensorboard application
# Install the Tensorboard application separately (part of tensorflow) to view them.
tensorboardX>=1.2

# Accessing files from S3 directly.
boto3

# REST interface for models
flask>=1.0.2
flask-cors>=3.0.7
gevent>=1.3.6

# Used by semantic parsing code to strip diacritics from unicode strings.
unidecode

# Used by semantic parsing code to parse SQL
parsimonious>=0.8.0

# Used by semantic parsing code to format and postprocess SQL
sqlparse>=0.2.4

# For text normalization
ftfy
word2number>=1.1

# To use the BERT model
pytorch-pretrained-bert>=0.6.0
git+git://github.com/huggingface/pytorch-transformers.git@a7b4cfe9194bf93c7044a42c9f1281260ce6279e

# For caching processed data
jsonpickle

#### ESSENTIAL LIBRARIES USED IN SCRIPTS ####

# Plot graphs for learning rate finder
matplotlib>=2.2.3

# Used for downloading datasets over HTTP
requests>=2.18

# progress bars in data cleaning scripts
tqdm>=4.19

# In SQuAD eval script, we use this to see if we likely have some tokenization problem.
editdistance

# For pretrained model weights
h5py

# For timezone utilities
pytz>=2017.3

# Reads Universal Dependencies files.
conllu==1.3.1

• Short description of package and use case(s):
  An open-source NLP research library, built on PyTorch

Folks who would also like to see this package in the supplementary channel, please thumbs-up so the developers can know which requests to prioritize.

[hartb]

I'm taking a look at this. Have a recipe (based on conda-forge + updates) that builds locally, but with a lot of unit tests failing at the moment.