Dependencies on PyPI needs to be updated.

Most users installs nltk through:

$ pip install -U nltk

or

$ conda install nltk

but somehow the PyPI requirements seems to be far lacking behind the ones in pip-req.txt.

[alvations]

I like @kmike suggestion too, having something like:

pip install nltk [plot] # Plotting + Tkinter + visual stuff included

and maybe:

pip install nltk [machine-learning] # All ML batteries included

and for lazy users:

pip install nltk [all] # All batteries included

---

On a related note, I think we need to also include the downloads in the setup.py code such that users won’t see an error when using “NLTK default” functions e.g. :

from nltk import sent_tokenize, word_tokenize, ne_chunk, pos_tag

Currently, now the users must download:

import download
nltk.download('punkt')
nltk.download('averaged_perceptron_tagger')

[kmike]

Adding all these packages to install_requires sounds like a maintenance nightmare - user must be able to build all these packages (many of them contain C/C++ extensions) in order to install NLTK - even if they are not needed at all for the user.

I don’t see them as required dependencies, they are optional dependencies which enables some of NLTK features. Currently the only required dependency is six - a small pure Python module. There is no reason for users to build gensim if they only want to use nltk tokenizer.

Nose, tox, coverage and pylint (and a few other packages which are their transitive dependencies) are not NLTK dependencies, they are packages required to run NLTK tests. They are not needed for the end users.