Skip to content

Latest commit

 

History

History
101 lines (76 loc) · 4.38 KB

README.md

File metadata and controls

101 lines (76 loc) · 4.38 KB
Raw

Prompts and Experiment Results Analysis App

This Flask application provides a web interface for entering prompts, uploading coded datasets, calculating metrics such as percentage of matches and Cohen's Kappa score, and downloading results as a CSV file. It is designed to facilitate the analysis of a series of experiments by comparing ground truth with inference data, while also tracking metadata for prompt engineering. This includes various versions of prompts, changes, and evaluation metrics. It will be used for the DEI summer undergraduate internship.

Example Usage

Input

PromptsOp Interface

Output

PromptsOp Interface

Features

  • Interactive Web Interface: Two main web pages for user interaction:
    • Home Page: Allows users to upload data and add original and revised prompts, as well as column names for ground truth and GPT-coded values for evaluation metrics (percent matched and IRR) calculation.
    • Results Page: Displays calculated metrics and provides options to download the data or go back to make changes.
  • Upload CSV Files: Users can upload CSV files to analyze experimental data.
  • Calculate Metrics: The app calculates the percentage of matches and Cohen's Kappa score for uploaded data.
  • Download Results: Users can download a summary of their results in CSV format.

Prerequisites

Before you can run this application, you need to have Python installed on your machine along with Flask and its dependencies. You will also need pandas and sklearn for data manipulation and metrics calculation.

Installation

  1. Clone the repository or download the source code.
  2. Navigate to the application directory.
  3. Set up a virtual nvironment (optional but recommended):
    • Install virtualenv if you haven't installed it yet: 
      pip install virtualenv
    • Create a virtual environment in your project folder: 
      python -m venv prompt_env
    • Activate the virtual environment:
      • On Windows: 
        prompt_env\Scripts\activate
      • On macOS and Linux: 
        source prompt_env/bin/activate
  4. Install required Python libraries by running: 
    pip install -r requirements.txt
  5. Create an environment file .env in the top directory of the project with your secret key:
    • Generate a secure secret key using Python or choose your own key: 
      import os
print(os.urandom(24).hex())
    • Open the newly created .env file and add the following line: 
      SECRET_KEY=your_generated_secret_key
    • Replace your_generated_secret_key with the key you generated. Ensure this key remains confidential and is not shared publicly. The secret key is needed for handling sessions in the app.

Usage

To start the application:

python3 promptApp.py

Once the application is running, navigate to http://127.0.0.1:5000/ in your web browser to access the app.

Uploading Data

  • Go to the home page and use the upload form to submit a CSV file.
  • Fill out the required fields including original and revised prompts, and the reasons for changes.
  • Specify the column names for ground truth and inference data in your CSV file.

Viewing and Downloading Results

  • After uploading the data, the application will display the calculated metrics on the results page.
  • You can view details about the prompts used and any changes made.
  • Use the provided link to download a CSV file of your results.
  • You can return to the home page to try another file by clicking the "Try another file" button, or to edit the previously entered information by clicking the "Go back" button.

Configuration

You can modify the column names used in the calculations by editing the ground_truth_column and inference_column variables within the upload_file function to match the column names in your CSV files.

Development

The application is set up with Flask's debug mode enabled to assist with development. Remember to disable debug mode in a production environment.

Contributing

Contributions to this project are welcome! Please fork the repository and submit pull requests with any enhancements or bug fixes. This is a playground for simple custom prompt ops.