FunkSVD Recommender System

This repository implements a recommendation system using FunkSVD (Funk Singular Value Decomposition). It predicts user-item ratings based on collaborative filtering, leveraging techniques like gradient descent optimization and bias adjustment.

Features

• FunkSVD Algorithm: A matrix factorization-based method for collaborative filtering.
• Hyperparameter Tuning: Includes options for tuning learning rate, regularization, and other key parameters.
• Mini-batch Gradient Descent: Efficient training with support for Adam optimizer.
• Train-Test Split: Custom utility to split data into training and testing sets.
• Prediction: Generates predictions for user-item pairs, including support for missing data.

---

File Structure

1. main.py

• The entry point for the system.
• Automates hyperparameter tuning using grid search.
• Splits data into train and test sets, trains the model, and evaluates performance using RMSE.
• Generates predictions for user-item pairs in targets.csv and saves results to output3.csv.

2. FunkSVD.py

• Contains the implementation of the FunkSVD algorithm:
  • Handles matrix factorization (P and Q matrices).
  • Supports bias adjustments for users and items.
  • Includes mini-batch gradient descent and the Adam optimizer for parameter updates.

3. TrainTestSplit.py

• Handles preprocessing and splitting of the dataset into training and testing sets.
• Reads user-item rating data from ratings.csv.

4. Adam.py

• Implements the Adam Optimizer, a popular optimization algorithm used in training machine learning models.
• Provides efficient updates for parameters during mini-batch gradient descent.

---

How to Use

1. Prerequisites

Ensure you have the following dependencies installed:

pip install numpy pandas

2. Prepare the Data

Place a ratings.csv file and a targets.csv file in the project directory. The file ratings.csv should have the format:

UserId:ItemId,Rating

and the targets.csv file should have the format:

UserId:ItemId

3. Run the Code

Execute main.py to:

1. Tune hyperparameters.
2. Train the FunkSVD model.
3. Generate predictions.

python3 main.py

4. Output

• During the training the Epoch and the current Loss are also printed in the stdout in the format:

Current_Epoch/Total_Epochs: Loss

• Predictions are printed in the stdout in the format:

UserId:ItemId,Rating

5. Hyperparameters

Key Hyperparameters for tuning the model

• epochs: Number of training iterations.
• lr: Learning rate for gradient descent.
• k: Number of latent factors.
• batch_size: Size of mini-batches.
• lamda: Regularization strength.
• test_size: Fraction of data used for testing.