PyTorch Regression Mutilayered Perceptron Model

This repository contains a PyTorch implementation of a regression Mutilayered Perceptron model designed for tabular data. It includes data loading, preprocessing, model definition, training, and evaluation functionalities.

Table of Contents

• Features
• Installation
• Usage
• Dataset Configuration
• Model Configuration
• Training Configuration
• Dependencies
• Logging
• Error Handling
• Metrics and Visualization
• Contributing
• License

Features

• Data Loading and Preprocessing:
  • Loads data from CSV files.
  • Supports chunked loading for large datasets.
  • Handles missing columns and empty datasets.
  • Implements various data scaling techniques (StandardScaler, MinMaxScaler, MaxAbsScaler, RobustScaler, Normalizer).
  • Adds noise to the data (Normal, Uniform, Poisson).
  • Allows scaling and noise parameters to be configured.
• Model Definition:
  • Defines a customizable neural network model with configurable hidden layers, dropout, and L1 regularization.
  • Includes batch normalization and ELU activation.
• Training and Evaluation:
  • Implements training and validation loops with error handling.
  • Supports early stopping based on validation loss.
  • Uses Adam optimizer with configurable learning rate, weight decay, and learning rate schedulers (StepLR, ReduceLROnPlateau).
  • Calculates and logs metrics (MSE, MAE, R-squared).
  • Collects and visualizes training and validation losses, MSE, MAE, and R-squared.
  • Plots residuals against predicted values.
• Device Management:
  • Automatically detects and uses available GPU (CUDA or MPS) or CPU.
• Logging:
  • Uses Python's logging module to log training and evaluation information.
  • Logs to both a file (learn_model.log) and the console.
• Error Handling:
  • Includes custom exceptions for dataset-related errors.
  • Comprehensive error handling throughout the code.

Installation

1. Clone the repository:

   git clone <repository_url>
   cd <repository_directory>

2. Install the required dependencies:

   pip install numpy pandas torch scikit-learn matplotlib

Usage

1. Prepare your dataset:

   • Place your CSV data file in the specified root directory.
   • Ensure the CSV file contains the columns specified in xcol and ycol within the DatasetConfig.

2. Configure the dataset:

   • Modify the DatasetConfig in the if __name__ == '__main__': block to match your dataset.
   • Set the root, csv_file, xcol, ycol, scaler_type, noise_type, noise_std, and scaling_factor as needed.

3. Configure the model and training:

   • Adjust the config dictionary in the if __name__ == '__main__': block to configure the model, optimizer, learning rate schedulers, and early stopping.

4. Run the script:

   python <your_script_name>.py

5. View the results:

   • The script will output training and evaluation metrics to the console and log file.
   • Plots of the training and validation losses, MSE, MAE, R-squared, and residuals will be displayed.

Dataset Configuration

The DatasetConfig dataclass allows you to configure the dataset:

@dataclass
class DatasetConfig:
    root: str
    xcol: list[str]
    ycol: list[str]
    scaler_type: ScalerType = ScalerType.STANDARD
    noise_type: NoiseType = NoiseType.NORMAL
    csv_file: str = 'dummy_data.csv'
    scaler: object = None
    noise_params: dict = None
    chunksize: int = None
    noise_std: float = 0.0
    scaling_factor: float = 1.0