From the Disrupt Hogwarts project (which consisted in developing from scratch a python library for logistic regression), I built a Flask API.
The API offers three functionalities:
- train: to train the model based on the initial data provided by 42.
- transfer: to train a new model for another school using "transfer learning". What it actually does, is initializing a new model with the one previously trained on the Hogwarts dataset, and updates the model to fit the data of the new school with a smaller learning rate. Transfer learning can help algorithms learn much quicker.
- predict: to make a prediction, i.e. compute P(y=house|X=x_i) for a given school.
Run git clone https://github.com/SachaIZADI/DisruptHogwarts_API.git in your shell.
Then mkvirtualenv --python=`which python3` hogwarts_api (You can escape the virtualenv by running `deactivate)
Then finally run `pip3 install requirements.txt. And You're up to play with the project.
Launch the local server, by running
python3 app.py
Output:
Server is now accessible on http://127.0.0.1:5000
The API can be run directly in the browser:
Or by using scripts such as test_api.py:
NB: Make sure that the BASE_URL variable is set upt to http://127.0.0.1:5000
I chosed to deploy the API on Heroku, mostly for time and money constraints. It is accessible on
https://hogwarts42-api-heroku.herokuapp.com
You can play with the model and make a couple predictions by running.
https://hogwarts42-api-heroku.herokuapp.com/predict?school=polytechnique&astronomy=-800&herbology=-2&ancient_runes=300
https://hogwarts42-api-heroku.herokuapp.com/predict?school=polytechnique&astronomy=800&herbology=2&ancient_runes=700
Alternatively, you can write scripts that directly query the API. Just change the BASE_URL value in test_api.py for instance.
.
Note to myself: Useful heroku guidelines and commands
A good article, and the classic commands
git push heroku master
heroku ps:scale web=1
heroku open
heroku logs