crowdnalysis-service

A service to compute and export the consensus for tasks in Citizen Science Project Builder via crowdnalysis.

The service acts as an intermediary between the C3S frontend of Citizen Science Project Builder and the underlying Pybossa API. The use case is as follows:

1. CS Project Builder user clicks the Export Results (in CSV or JSON format) button on the C3S frontend;
2. The request is forwarded to the crowdnalysis-service;
3. The service
   • Calls Pybossa API to extract task, task_run and result data,
   • Computes the consensus on tasks for each question asked to the crowd by using crowdnalysis with the given consensus model (Dawid-Skene by default),
   • Creates a CSV or a JSON file for each consensus depending on user's request,
   • Sends the consensus and original result files back to the C3S in a .zip file;
4. The user downloads the .zip file without leaving the C3S frontend in any of the above steps.

If consensus does not apply to the project (e.g., there is no classification task) or an error occurs during consensus computation, only the result files are returned. Log messages are sent to the stderr.

Starting the service

The service is basically a Flask application running on a Gunicorn WSGI server and listening on the 5000 port. After git cloning the repo, start the service:

As a standalone app

source bin/init.sh && bin/boot.sh 

As a docker container

First, build the docker image:

docker build --tag crowdnalysis-service .

Then, run the container (in detached mode):

docker run -d -p 5000:5000 --env-file service.env crowdnalysis-service

If you run the service on the same host with Pybossa, use the following command instead:

docker run -d -p 5000:5000 --env-file service.env --add-host host.docker.internal:host-gateway crowdnalysis-service

Customization

• Edit the service.env file to configure the PORT that the service listens on and, if you start the service as a container, set the -p option in docker run accordingly.
• To view DEBUG-level log messages, set the related environment variable to 1 in the same file.

When the service is started by docker-compose within the Pybossa multi-container set-up, the environment variables are read from the etc/crowdnalyis-service.env file instead.

Troubleshoot

Docker image

• While building the docker image, if you experience the 'internal compiler error: Killed (program cc1plus)' error during the installation of CmdStan, increase the memory dedicated to your Docker engine, and retry building.