Media Cloud Word Embeddings Server

A micro-service to support analyzing words based on models of word embeddings (aka. "word2vec").

Dev Installation

1. Install python 3.6.5.
2. Run pip install -r requirements.txt to install dependencies
3. Run python download-google-news-model.py to download the google news model file.

Developing

We develop with PyCharm.

Configuration happens via environment variables. We use python-dotenv to manage this on locall dev machines. Make a .env file and define the following in it:

• SENTRY_DSN - optional url for Sentry logging
• SECRET_KEY - for Flask
• LOG_LEVEL - DEBUG, INFO, etc.
• MEDIA_CLOUD_API_KEY - your Media Cloud API key

Running

Two options:

1. Development: run python run.py to test it out
2. Production-like: run ./run.sh to run it with gunicorn

You can then hit the local homepage to try it out from a simple web-testing harness: http://localhost:8000

Or you can test that with something like this (the first request takes a while to load the giant model into memory):

import requests
response = requests.post("http://localhost:8000/api/v2/google-news2d.json",
                         data = {'words[]':['apples', 'bananas', 'three']})
print response.json()

Deploying

This is configured to deploy as a Heroku buildpack to dokku.

You'll need to do something like this to set the required environment variables:

dokku config:set word-embeddings SECRET_KEY=oiwajj243josadjoi SENTRY_DSN=https://THING1:THING2@sentry.io/THING3 MEDIA_CLOUD_API_KEY=MY_AWESOME_KEY

Releasing

1. Update the semantic version number in server/__init.py__
2. Tag the repository with that number, like v4.5.2
3. Push it to the server, like git push dokku v4.5.2:master