Here’s the updated README.md with attribution to the Udemy course:

---

Boltzmann Machines for Recommendation Systems

Overview

This project implements a Restricted Boltzmann Machine (RBM) for movie recommendation. The RBM learns hidden patterns in user-movie interaction data, enabling it to predict unseen user preferences.

Note: This implementation is based on the course materials from the Udemy Deep Learning A-Z™ Course.

---

Dataset

The project uses the MovieLens Dataset:

• movies.dat: Contains movie details.
• users.dat: Contains user information.
• ratings.dat: Contains user ratings for movies.
• Training and Test Splits:
  • u1.base: Training dataset.
  • u1.test: Test dataset.

---

Preprocessing

1. Data Conversion:

   • The user-movie interactions are converted into a matrix where:
     • Rows: Users.
     • Columns: Movies.
     • Values: Ratings (1 to 5) or 0 (not rated).
   • Ratings are then converted into binary values:
     • 1: Liked (ratings 3, 4, or 5).
     • 0: Disliked (ratings 1 or 2).
     • -1: Not rated.

2. PyTorch Tensor Conversion:

   • The data is converted into PyTorch tensors for model processing.

---

RBM Architecture

The RBM consists of:

• Visible Layer:
  • Represents the input data (movie ratings).
  • One node per movie.
• Hidden Layer:
  • Represents latent features, such as user preferences.
  • Configured with 100 hidden nodes in this implementation.
• Weights and Biases:
  • Weights (W): Connect visible and hidden layers.
  • Biases:
    • a: Hidden layer biases.
    • b: Visible layer biases.

RBM Methods

1. sample_h:

   • Computes hidden probabilities given visible nodes.
   • Samples binary activations for the hidden layer.

2. sample_v:

   • Computes visible probabilities given hidden nodes.
   • Samples binary activations for the visible layer.

3. train:

   • Updates weights and biases using Contrastive Divergence.

---

Training

• Parameters:
  • Number of epochs: 10.
  • Batch size: 100.
  • Contrastive Divergence: 10 steps.
• Procedure:
  • For each epoch:
    • Perform forward pass (positive phase) to compute hidden probabilities.
    • Perform reconstruction (negative phase) with Contrastive Divergence.
    • Update weights and biases using the difference between the positive and negative phases.
• Loss Function:
  • Mean absolute error between the original and reconstructed visible nodes.

---

Testing

• Procedure:
  • Predict ratings for each user in the test set using the trained RBM.
  • Compute mean absolute error between predicted and actual ratings for rated movies.
• Output:
  • Final test loss (mean absolute error across all users).

---

Key Results

• The RBM learns to reconstruct user preferences based on movie ratings.
• Test loss provides a measure of how well the RBM predicts unseen data.

---

How to Run

1. Clone the repository:

   git clone https://github.com/your-repo-name.git
   cd your-repo-name

2. Install dependencies:

   pip install numpy pandas torch

3. Place the dataset files (movies.dat, users.dat, ratings.dat, u1.base, u1.test) in the working directory.

4. Run the script:

   python Boltzmann_Machines.py

---

Directory Structure

├── Boltzmann_Machines.py       # Main implementation
├── README.md                   # Project documentation
├── ml-1m/                      # Dataset directory
│   ├── movies.dat
│   ├── users.dat
│   ├── ratings.dat
├── ml-100k/                    # Dataset splits
│   ├── u1.base
│   ├── u1.test

---

Dependencies

• Python 3.x
• PyTorch
• NumPy
• Pandas

---

Acknowledgment

This project is based on the implementation from the Udemy Deep Learning A-Z™ Course.