LabelForward

A web application derived from the CovidState project. It provides an interface for labeling text data, using active learning workers to speed up the labeling process. It can be build and run using docker-compose, which will start up a docker environment with microservices for frontend, backend, database, active learning & prediction, data downloads and a database setup.

Prerequisites

To build this project, only Docker must be installed on your machine. You also need to make sure that port 3050 is not already in use on your machine.

Build

To build the project, open a terminal and navigate to the directory project directory. After navigating to the directory, execute the following command to build the development version:

docker-compose build 

To build the production version, execute the following command:

docker-compose -f docker-compose-prod.yml build

Usage

To run the project in development mode, execute the following command:

docker-compose up

To run the poject in production mode, execute the following command:

docker-compose -f docker-compose-prod.yml up

You can now access the application at http://localhost:3050

To shutdown the application, enter Ctrl + C in the terminal you started the application with. Alternatively, navigate to the project directory and execute the following command:

docker-compose down

Combine Build & Run

The build and run process can also be combined. For development mode:

docker-compose up --build

For production mode:

docker-compose -f docker-compose-prod.yml up --build

Troubleshooting

Sometimes the project might not update correctly when being rebuilt with changes in the services. Instead of pulling the project again and rebuild it from scratch, try executing the following command in the main directory:

docker-compose down --volumes && docker-compose up --build

Add your own data

The aforementioned process builds the project with example data, a news headline dataset with with labels link. To use it with your own data, you need to make a few adjustments.

1. Prepare your data

The data and the labels must be provided as a JSON file. The file needs the format of a pandas DataFrame as json. You also need to define a list of accessors (eg. email addresses of annotators) in the same format. See the db_helper/exampledata for examples. Note that you need to locate your data, samples and accessors somewhere within the db_helper folder.

2. Adjust the docker-compose files

In the docker-compose files, you need to adjust the paths DATA_PATH, LABEL_PATH AND ACCESSOR_PATH to your JSON file paths.