regression-using-pytorch-tutorial

A tutorial on building different regression models using the PyTorch modules

Weather Temperature Regression Using PyTorch

A machine learning project that implements neural network regression models to predict temperature based on weather features using PyTorch.

📊 Dataset

Source: Weather Dataset from Kaggle

Dataset Description: This project uses a comprehensive weather dataset containing 8,784 hourly weather observations with multiple meteorological features.

Features Used:

• Dew Point Temperature (°C): Primary predictor variable
• Relative Humidity (%): Secondary predictor variable

Target Variable:

• Temperature (°C): The variable we aim to predict

Data Preprocessing:

• Removed non-numeric columns (Date/Time, Weather)
• Selected the two most correlated features based on exploratory data analysis
• Split data into 80% training and 20% testing sets

🔍 Exploratory Data Analysis

The project includes comprehensive EDA with:

• 2D scatter plots analyzing individual feature relationships with temperature
• 3D visualizations exploring combined feature effects
• Correlation analysis to identify the strongest predictors

Key findings:

• Dew Point Temperature shows the strongest linear relationship with Temperature (R ≈ 1.0)
• Relative Humidity has a moderate negative correlation with Temperature
• Other features (Wind Speed, Visibility, Pressure) show weaker relationships

🏗️ Model Architecture

Model 1: ReLU Activation

Input (2) → Linear(20) → ReLU → Linear(10) → ReLU → Linear(1) → Output

• Activation Function: ReLU
• Hidden Layers: 2 layers (20, 10 neurons)
• Optimizer: Stochastic Gradient Descent (SGD)
• Learning Rate: 0.0001

Model 2: Sigmoid Activation

Input (2) → Linear(15) → Sigmoid → Linear(8) → Sigmoid → Linear(1) → Output

• Activation Function: Sigmoid
• Hidden Layers: 2 layers (15, 8 neurons)
• Optimizer: Adam
• Learning Rate: 0.001

📈 Results

Both models achieved exceptional performance:

Metric	Model 1 (ReLU)	Model 2 (Sigmoid)
MAE	0.1432	0.1432
MSE	0.0434	0.0434
RMSE	0.2082	0.2082
R²	0.9997	0.9997

🚀 Getting Started

Prerequisites

pip install torch torchvision
pip install pandas numpy matplotlib
pip install scikit-learn

Installation

1. Clone the repository:

git clone https://github.com/yourusername/weather-temperature-regression.git
cd weather-temperature-regression

2. Download the dataset:

   • Place Weather Data.csv in the project directory
   • Or download from the Kaggle link provided above

3. Run the Jupyter notebook:

Jupyter Notebook regression-with-pythrch.ipynb

🔧 Usage

The notebook is structured in the following sections:

1. Data Loading and Exploration
2. Exploratory Data Analysis with Visualizations
3. Data Preprocessing and Feature Selection
4. Model 1 Implementation (ReLU)
5. Model 2 Implementation (Sigmoid)
6. Performance Evaluation and Comparison
7. Analysis and Recommendations

📊 Key Insights

• Strong Linear Relationship: Dew point temperature is an excellent predictor of air temperature
• Model Performance: Both activation functions (ReLU and Sigmoid) performed equally well for this regression task
• Feature Engineering: Careful feature selection based on correlation analysis significantly improved model performance
• Generalization: High R² score (99.97%) indicates excellent model fit without apparent overfitting

🤖 Technical Implementation

Neural Network Features:

• PyTorch Framework: Built using PyTorch's nn.Module
• Custom Model Classes: Implemented two distinct architectures
• Multiple Optimizers: Compared SGD vs Adam optimization
• Comprehensive Metrics: MAE, MSE, RMSE, and R² evaluation

Training Configuration:

• Epochs: 100,000 for both models
• Loss Function: Mean Squared Error (MSE)
• Data Split: 80/20 train-test split with random_state=42

📝 Recommendations

Based on the analysis:

1. Model Choice: Either model is suitable for this task, given identical performance
2. Activation Function: ReLU might be preferred for scalability to larger networks
3. Feature Selection: Dew point temperature alone could potentially achieve similar results
4. Production Deployment: Consider the ReLU model for computational efficiency

🎓 Academic Context

This project was completed as part of:

• Course: ANLY-6500: Advanced Data Analytics
• Module: 3 - Regression Using PyTorch

📧 Contact

Author: Ebube Ndubuisi

For questions or collaborations, please open an issue or submit a pull request.

📄 License

This project is available under the MIT License. See LICENSE file for details.

---

This project demonstrates the application of deep learning techniques for weather prediction, showcasing the effectiveness of neural networks in capturing meteorological relationships.