A framework around prodigy to let end users start and stop the interface, trigger training, etc. - for small group collab.
-
Make sure you have an active prodigy license
-
Create a .env file in the build directory (or edit docker-compose.yml in that location), containing:
COMPOSE_PROJECT_NAME=concept_miner_elastic KIBANA_SERVER_PUBLICBASEURL=https://<your_base_url>/kibana TZ=<your_timezone> DEBUG=false PRODIGY_KEY=<your_prodigy_key>
-
Run
docker compose up
-
The server should now be running on http://127.0.0.1:8000/ (you can change this port in ./conf/nginx.conf)
- Visit http://<your_ip>:8000/
- From the Home tab, upload your PDF or TXT documents for training
- Navigate to the Annotating tab, and perform your prodigy annotations
- Once you have completed a number of annotations, return to 'Home' and click 'Go' next to 'Stop annotating (Start training)'
- Results will be on the 'Training Log' tab, and your model and annotations will be in the ./output folder (or as configured in app.conf)