EEG Oddball Classification Experiments

A modular Python framework for EEG classification experiments using P3 and Active Visual Oddball (AVO) datasets. This codebase supports both individual subject training and cross-dataset generalization experiments with multiple neural network architectures.

Installation

1. Clone the repository:

   git clone <repository-url>
   cd EEG_experiments

2. Install dependencies:

   pip install -r requirements.txt

3. Set up data paths in config.py:

   P3_DATA_DIR = "path/to/your/P3/data"
   AVO_DATA_DIR = "path/to/your/AVO/data"

Quick Start

Basic Usage

python main.py

The experiment configuration is controlled through config.py. Here are the key parameters:

Configuration Options

1. Dataset Selection

# Single dataset experiments
data_dir = P3_DATA_DIR  
dataset = 'P3 Raw Data BIDS-Compatible'
use_combined_datasets = False

# Combined dataset experiments  
use_combined_datasets = True  # Uses both P3 and AVO

2. Training Strategy

# Pooled training: All subjects' data combined into one model
separate_subject_classification = False

# Individual training: Each subject gets their own model  
separate_subject_classification = True

3. Electrode Configuration

# Use common electrodes across datasets (recommended for combined experiments)
electrode_list = 'common'  

# Use all available electrodes for single dataset
electrode_list = 'all'

4. Classifier Selection

# Deep learning approach (recommended)
classifier = 'ShallowFBCSPNet'

# Traditional machine learning
classifier = 'lda'  # Linear Discriminant Analysis

5. Custom Hyperparameters

# Training configuration
BATCH_SIZE = 64
MAX_EPOCHS = 100  
LEARNING_RATE = 0.001
TRAIN_SIZE = 0.6 
VAL_SIZE = 0.2    
TEST_SIZE = 0.2   

# Reproducibility
seeds = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]  # Multiple seeds for robust results

Log Files

Detailed logs are saved in the ./log/ directory with timestamps and configuration:

log/Combined_clf-ShallowFBCSPNet_sep-False_el-common_results_20241201_143022.log

Adding New Datasets

1. Add preprocessing logic in preprocessor.py
2. Update constants in constants.py
3. Modify experiment logic in experiment.py
4. Update configuration in config.py

Adding New Models

1. Add model creation in models.py:

   def create_model(n_channels, is_lda=False):
       if is_lda:
           return LinearDiscriminantAnalysis()
       # Add your custom model here
       return YourCustomModel(n_channels)

2. Update config.py:

   classifier = 'YourCustomModel'

Code Structure

├── config.py           # All experiment parameters and hyperparameters
├── constants.py        # Channel names and event code definitions  
├── preprocessor.py     # OddballPreprocessor class for EEG data processing
├── models.py          # Model creation, training, and evaluation functions
├── utils.py           # Data loading, statistics, and utility functions
├── experiment.py      # Core experiment logic and training procedures
├── main.py            # Main entry point and experiment orchestration
├── experiment_logger.py # Logging and result tracking utilities
└── requirements.txt   # Python package dependencies