Skip to content

Turing Code Classifier - GUI Documentation

The Graphical User Interface (GUI) for the Turing Code Classifier is built with Gradio and integrated into FastAPI. This interface allows developers and stakeholders to interact with the model visually and provide human-in-the-loop feedback.

Overview

The GUI serves two primary purposes:

  • Real-time Inference: Allows users to input code comments and receive immediate classification predictions from the model
  • Data Collection: Enables users to correct model predictions, which are saved locally to create a dataset for future fine-tuning

Technical Architecture

The GUI is mounted onto the main FastAPI backend.

  • Framework: Gradio
  • Integration: Mounted via gr.mount_gradio_app within FastAPI
  • Location: Accessible at the /gradio endpoint
  • Source Code:
  • turing/api/demo.py: Contains the UI layout, custom CSS, and event logic
  • turing/api/app.py: Handles the mounting of the interface to the API

How to Run

Since the GUI is integrated into the API, you simply need to start the main application server.

  1. Navigate to the project root
  2. Start the server using Uvicorn:
uvicorn turing.api.app:app --reload
  1. Access the interface in your browser:
  2. Direct Link: http://127.0.0.1:8000/gradio
  3. Via API Docs: Click the indicated logo at http://127.0.0.1:8000/docs

User Guide

1. Classifying Comments

  1. Input Source: Paste the code comment (e.g., // TODO: Fix this hack) into the large text box
  2. Language: Select the source programming language (java, python, or pharo) from the dropdown menu
  3. Classify: Click the Classify button
  4. Result: The predicted category will appear in the "Analysis Result" box on the right

Note

Syntax Validation: The system automatically checks if the input syntax matches the selected language. If you try to classify a Java comment (e.g., starting with //) while "Python" is selected, the system will return an error message instead of a prediction.

2. Submitting Feedback

If the model's prediction is incorrect, you can help improve it:

  1. Look at the "Help Improve the Model" section below the result
  2. Correct Label: Select the actual correct category from the dropdown menu
  3. Note: The options in this menu update dynamically based on the selected programming language
  4. Submit: Click Save Feedback. A confirmation message will appear

Feedback Data Storage

User feedback is persisted to a local CSV file. This file is intended to be used to analyze model errors and retrain the model.

File Path: reports/feedback/feedback_data.csv

Data Schema:

Column Name Description Example
Timestamp Date and time of submission (YYYY-MM-DD HH:MM:SS) 2025-12-12 14:30:00
Input_Text The code comment provided by the user "# This is a setup"
Language The programming language context python
Model_Prediction The category originally predicted by the AI Summary
User_Correction The correct category selected by the user Usage

Note

The CSV file acts as an append-only log. It is automatically created if it does not exist.

UI Customization

  • Dark Mode: Fully supported. The UI automatically adapts to system preferences or browser settings
  • Assets: The header displays a custom SVG logo loaded from reports/figures/logo_header.svg
  • Dev Notes: A dynamic ticker in the header displays "developer thoughts"