Proxy Wars

1. Introduction

The Proxy Wars Tool helps data scientists identify and mitigate bias-inducing proxy variables within datasets. Proxy variables are attributes that correlate with sensitive variables (e.g., gender, race, age) and may unintentionally introduce bias into machine learning models.

Key Features:

• CSV Dataset Upload: Analyze numerical data from uploaded files.
• Algorithm Selection: Choose from Correlation Analysis, FACET, or Association Rule Mining (ARM) to identify proxy variables.
• Dataset Filtering: Refine datasets using random sampling or SQL-based filters.
• Dark Mode Support: Toggle between light and dark themes for a better user experience.
• Results Visualization: Display outputs in dynamic tables with sorting capabilities.

---

2. Getting Started

System Requirements

• Operating System: Windows, macOS, or Linux

Dependencies:

• Node.js: v18.16.0 or higher
• Python: v3.11 or higher
• Flask: v2.3.3
• React: v18.2.0

Installation

Install Dependencies:

1. Navigate to the /frontend folder and run:

   npm install

2. Navigate to the /backend folder and run:

   pip install -r requirements.txt

Run the Application:

Using Docker Container

1. If not already installed, download Docker Desktop from here.

2. Verify Docker installation:

   docker --version

3. Navigate to your project folder:

   cd path/to/project

4. Build the Docker container:

   docker-compose build

5. Start the Docker container:

   docker-compose up

6. Access the tool in Chrome: Open http://localhost:3000.

---

Running on Terminals

Open 2 terminals: one for the frontend and one for the backend.

Backend Setup:

1. Navigate to the backend folder:

   cd backend

2. Install Python dependencies:

   pip install -r requirements.txt

3. Run the Flask app:

   python src/controllers/app.py

Frontend Setup:

1. Navigate to the frontend folder:

   cd frontend

2. Install frontend dependencies:

   npm install

3. Start the development server:

   npm start

---

3. Features and Use Cases

1. Uploading a Dataset

• Click the Upload Dataset button.
• Select a valid .csv file (only numerical columns).
• Click Upload.
• A confirmation message will appear, and the dataset columns will be listed.

---

2. Switching Between Dark Mode and Light Mode

• Locate the Dark Mode Toggle in the top-right corner of the interface.
• Click to switch between Light Mode (default) and Dark Mode.

---

3. Algorithm Selection

• From the dropdown menu, choose an algorithm:

  • Correlation Analysis: Calculates relationships between variables.
  • FACET: Detects redundancy using feature selection.
  • Association Rule Mining (ARM): Generates association rules.

• If selecting FACET, specify a Target Variable.

---

4. Dataset Filtering Options

• Complete Dataset: Use the full dataset.
• Random Sampling:
  • Specify a percentage (e.g., 50%).
  • Enter a Random Seed for reproducibility.
• SQL Filter: Enter a filter condition (e.g., age > 30 AND income < 50000).

---

5. Viewing Results

• Click Results to run the analysis.

• The results are displayed in a table:

  • Correlation Analysis: Displays Pearson, Kendall, and Spearman coefficients.
  • FACET: Displays redundancy metrics.
  • ARM: Displays support, confidence, and lift values.

• Sort results by clicking on the column headers to organize by any metric.

---

6. Dark Mode Results Visualization

• All tables and UI elements will adapt to Dark Mode when enabled.

---

7. Sorting Results

• Use dropdowns to:
  • Select a column for sorting.
  • Choose a metric (e.g., Pearson coefficient for Correlation Analysis).
• Click the Sort button to organize results.

---

4. Troubleshooting

Common Issues:

• No File Uploaded: Ensure you have selected a valid .csv file.
• Target Variable Missing: Choose a target variable if using FACET.
• Sensitive Variables Not Set: Ensure you have selected sensitive variables for analysis.

---

5. Appendix

Algorithm Details:

• Correlation Analysis: Calculates relationships between variables using Pearson, Kendall, and Spearman coefficients.
• FACET: Detects redundancy using Random Forest feature selection.
• Association Rule Mining (ARM): Identifies patterns and associations with metrics like support, confidence, and lift.

Example Datasets:

• Titanic Dataset: Analyze survival likelihood based on various features.
• Census Data: Explore income-related proxies, such as education and occupation.

---