flask-label-ui

Starter code for simple multi-class labeling with flask.

Getting Started:

If you're just going to use this locally, you can just check out the code and modify it. If you want to back that up, I recommend creating your own repo based on this, by clicking 'Use this template'.

If you plan to check in your dataset, think about the legal implications of sharing your data!. Make your repo private if your data is <100MB and you plan on checking it into git.

Get the dependencies you need:

pip install -r requirements.txt # flask & sqlalchemy

Running the Server:

As one line:

export FLASK_ENV=development; export FLASK_APP=server.py; flask run

As separate steps (in run.sh)

export FLASK_ENV=development
export FLASK_APP=server.py
flask run

Then navigate to localhost:5000 to view it locally.

Customizing:

1. Open dataset.py:
   1. Change DEFAULT_BUTTONS to whatever classes you expect to have.
   2. Change load_examples to point at your file; example loaders for CSV and JSONL are provided.
      • Feel free to downsample your data to ~3000 or so at this point (You're unlikely to label that many!).
   3. Create any fields you might want on Example -- by default they're all shoved into features.
2. Change templates/example.j2 to render your particular data in a user-friendly way. Each row is currently formatted as JSON.
   • You can read more about Jinja2 templates at their documentation site.

Running it publicly (to share with a partner)

1. Get it to run locally first.
2. (CS451 students) Add me as a collaborator to your repo.
3. (Everyone else) Find your credit-card and seek hosting! Digital Ocean, Vultr, Heroku (need to switch to postgres), etc.