League of Legends Match Predictor

Overview

This project implements a logistic regression model using PyTorch to predict the outcomes of League of Legends matches. It demonstrates comprehensive data loading, preprocessing, model training, hyperparameter tuning, evaluation, and feature importance analysis.

Dataset

The dataset used in this project:

• league_of_legends.csv

Project Steps

Step 1: Data Loading and Preprocessing

• Load dataset using pandas
• Split data into training and testing sets (80%-20%)
• Standardize features using StandardScaler
• Convert data into PyTorch tensors

Step 2: Logistic Regression Model

• Define logistic regression model architecture with PyTorch
• Initialize model, loss function (Binary Cross Entropy Loss), and optimizer (Stochastic Gradient Descent)

Step 3: Training the Model

• Train the logistic regression model for multiple epochs
• Evaluate training performance periodically

Step 4: Model Optimization

• Apply L2 regularization (weight decay) to prevent overfitting
• Retrain the model and evaluate performance improvements

Step 5: Model Evaluation and Visualization

• Compute confusion matrix and classification report
• Plot ROC curve and calculate AUC to assess model quality

Step 6: Save and Load the Model

• Demonstrate techniques to save (torch.save) and load (torch.load) PyTorch models
• Evaluate the loaded model to confirm consistent performance

Step 7: Hyperparameter Tuning

• Perform tuning to identify the best learning rate from [0.01, 0.05, 0.1]
• Evaluate test accuracy for each learning rate and select the best

Step 8: Feature Importance

• Extract and visualize model weights to interpret feature significance

Required Libraries

pip install pandas scikit-learn torch matplotlib

License

This project is licensed under the MIT License.