AI Assignment Grader

This project contains two Python scripts for training and testing a model that grades assignments based on provided guidance and a set of pre-graded assignments.

Files

• train.py: This script loads the assignment guidance, assignment texts, and grades, then trains a BERT model for sequence classification to predict the grade based on the guidance and assignment text.

• test.py: This script loads a trained model and uses it to predict the grade of a test assignment.

• guidance.txt: The guidance provided for the assignments.

• test_assignment.txt: A test assignment for which to predict the grade.

• assignments/: A directory that contains the assignment texts and their grades.

Usage

1. Place your assignment guidance in the guidance.txt file.

2. Place your pre-graded assignments in the assignments/ directory. Each assignment should have a .txt file for the assignment text and a _grade.txt file for the grade (which should be 'U', 'Pass', 'Merit', or 'Distinction').

3. Run the train.py script to train the model:

   python train.py

4. Place the text of the assignment you want to grade in the test_assignment.txt file.

5. Run the test.py script to predict the grade of the test assignment:

   python test.py

The grade of the test assignment will be printed to the console.

Requirements

• Python 3.7 or later
• PyTorch
• Transformers
• Other dependencies can be installed with pip install -r requirements.txt.

Disclaimer

This model is not guaranteed to provide accurate grades and should be used as a tool to assist in grading rather than a replacement for human grading. This currently only works with assignments with 400ish words, so not ideal for anything above and would give very inaccurate results.

For a model that can cope with a higher word count visit:

https://github.com/Cossy179/Grade-Calculator-LongForm