python + django + scikit-learn + mecab (1) on heroku

How to use python scikit-learn + mecab on heroku.

heroku has heroku-buildpack-mecab for handling mecab in Ruby, but not for python. So use heroku-buildpack-linuxbrew to brew install mecab.

Also, heroku-buildpack-python cannot install libraries that require a c compiler, such as scipy and scikit-learn. Instead, install miniconda instead of python with heroku-buildpack-conda to install these packages.

In order to handle these multiple buildpacks, you need to create an app with heroku-buildpack-multi.

First in the local repository for heroku

$ git init
$ heroku create --buildpack https://github.com/heroku/heroku-buildpack-multi

Create an app with.

Then create .buildpacks in your local repository

.buildpacks

https://github.com/kennethreitz/conda-buildpack.git
https://github.com/sunny4381/heroku-buildpack-linuxbrew.git

Write and save. buildpack-multi will install the buildpack listed in this file.

Create a .cellar in the same directory

.cellar

mecab
mecab-ipadic

Write and save. buildpack-linuxbrew will install the applications listed in this file.

The libraries you want to install with pip and conda are If you describe it in requirements.txt and conda-requirements.txt, it will be installed without permission.

$ pip freeze > requirements.txt

Or

$ conda list > requirements.txt

** Note that mecab-python cannot be installed with pip ** (maybe I should increase the version of pip ??)

sh: 1: mecab-config:I get a not found error

 So mecab-python is installed manually on heroku (quite brute force ...)
 In preparation for that, in the local repository

```shell-session
$ curl -O https://mecab.googlecode.com/files/mecab-python-0.996.tar.gz
$ tar zxfv mecab-python-0.996.tar.gz
$ rm https://mecab.googlecode.com/files/mecab-python-0.996.tar.gz

Extract mecab-python.

Also change 'mecab-config' in setpu.py in mecab-python-0.996 to `` `'/app/.linuxbrew/bin/mecab-config'```. (Reference source: http://qiita.com/saicologic/items/ab70e14f7e2ec2ee0b4d)

After all these things are done, deploy with bash.

$ heroku config:add LD_LIBRARY_PATH=/app/.linuxbrew/lib
$ heroku config:set MECAB_PATH=/app/.linuxbrew/lib/libmecab.so
$ git add .
$ git commit -m 'initial'
$ git push heroku master

After deploying, manually install mecab-python on heroku.

$ heroku run bash
~/ cd mecab-python-0.996
~/ python setup.py build
~/ python setup.py install

OK ... Finally, start python on heroku and check if import MeCab can be done.

Now you can use python + scikit-learn + mecab on heroku, but the file size is messed up ...

I feel that the free 300M is pretty tough ... (Reference source: http://qiita.com/shouta-dev/items/cd538a77f2b729333025)