EST_Classifier

Burokratt-Classifier-Prototype-Rootcode

This repository serves as a prototype for the Burokratt-Classifier. The following sections provide details:

1. How to Run the Prototype

Ensure that Docker and Git are properly installed and configured. Then, clone the GitHub repository using the following command:

git clone https://github.com/rootcodelabs/Burokratt-Classifier-Prototype-Rootcode.git

Navigate to the cloned repository directory.

Build and run the Docker containers using the following Docker command:

docker-compose up --build

And then you can view the web app by going to http://localhost:3000/

To view logs, use the command:

docker-compose logs -f

To stop the containers, use the command:

docker-compose down

After running the Docker containers, access the web application using the provided web address.

2. What is Included in the Prototype

The prototype consists of the following main sections accessible from the sidebar:

1. Dataset:

   • View current datasets and import new datasets in JSON or CSV format.
   • Expectations for unlabeled JSON files: ["Example1....","Example2....","Example3....",....]
   • Expectations for labeled JSON files:

     {
         "class1":["Example1....","Example2....","Example3....",....],
         "class2":["Example4....","Example5....","Example6....",....],
         "class3":["Example7....","Example8....","Example9....",....]
     }

   • Expectations for unlabeled CSV files: CSV file with one column containing example data for a class.

2. Classes:

   • Add, view, or delete classes.
   • Add more data to existing classes.

3. Models:

   • View accuracy, F1 score, and precision of existing models for each class.
   • Start creating a model by selecting a class and base NLP model (Bert, Albert, and XLNet).
   • Note: Models are trained for only one epoch due to prototype limitations.

4. Test:

   • Test a trained model by selecting the model and passing string input.

3. Special Points to Consider

• YAML files and labeled CSV files are not considered in the prototype.
• All models are set to train for one epoch to save time and resources, but this is configurable through code.