Here’s a concise and informative README.md for your script:

---

🧪 Multi-Armed Bandit Simulation for Product Recommendation Algorithms

This project simulates user interactions with product recommendation algorithms using multi-armed bandit strategies—Thompson Sampling, Upper Confidence Bound (UCB), and Random Selection—to evaluate which algorithm drives more user engagement (clicks).

📌 Purpose

The goal is to simulate and analyze how different recommendation algorithms perform in driving user selections from various product divisions. It seeks to answer:

• Does the algorithm type affect user decisions?
• Is there a statistically significant difference in engagement between algorithms?

🏗️ Architecture Overview

The simulation comprises the following stages:

1. Data Initialization

• Reads product data from Items_without_duplicates.csv.
• Extracts product divisions (categories).
• Creates a list of User objects from a user_data class, initializing internal tracking for each algorithm.

2. Trial Simulations

For each algorithm:

• Runs multiple trials for each user.
• In each trial:
  • Recommends a division based on the algorithm.
  • Uses bias-based acceptance to simulate user behavior.
  • Updates internal tracking stats (clicks, impressions).

3. Statistical Evaluation

• Performs a Chi-Squared Test to assess if user engagement varies by algorithm.
• Conducts a T-Test to determine if there's a statistically significant difference in click rates between UCB and Thompson Sampling.

📊 Output

• Total clicks generated by each algorithm.
• True/False indicating whether the algorithm type significantly impacts user behavior.
• True/False indicating whether there's a significant difference between Thompson and UCB.

🔧 Key Components

• initialize_user_collection(...): Initializes users for each algorithm group.
• generate_interaction_data_from_multiple_runs(...): Simulates user trials and updates user state.
• check_if_algorithm_type_affects_user_decisions(...): Chi-squared test for independence.
• check_for_significant_difference_between_clicks_due_to_algo_type(...): T-test for difference of means.

📁 Dependencies

Make sure to have the following:

• Items_without_duplicates.csv (Product data with at least dept_name and prod_name columns)
• user_data class implementation
• Utility functions in utils.py:
  • recommendation_by_thompson
  • recommendation_by_ucb
  • recommendation_by_random
  • get_division_probabilities
  • chi_squared_test_for_independence
  • t_test_for_difference_of_means