How universumx/universum works
Project Type and Core Purpose
• The project is a research-driven, real-time predictive system that integrates advanced data collection from an 8-channel BCI (EEG) with sophisticated preprocessing and machine learning techniques (including CNNs, Autoencoders, LSTM networks, and Transformers) to predict lumbar tilt perception.
• It is essentially a hybrid system with clear separation of data acquisition, preprocessing, and model development components.
Analyzing the File Structure
• The project is divided into two main directories: “DataCollection” and “ModelDevelopment.”
• The DataCollection directory handles the ingestion, organization, and preliminary processing of raw data. It contains:
• Assets (audio files that may be used for synchronization or stimulus purposes).
• A deep hierarchical “data” folder structure where raw EEG, accelerometer, and action data are stored, along with supporting files like “subject_info.csv.”
• Modules and scripts (e.g., “data_collection.py,” “data_visualization.py,” and additional modules for local storage or influx data integration).
• Supplementary components such as a GUI (“gui.py”) and procedural documentation (“procedure.md”).
• The ModelDevelopment directory focuses on building the predictive models and processing pipelines. It includes:
• Scripts for preprocessing (e.g., “preprocessing.py”), real-time processing (e.g., “realtime.py”), and analyses (e.g., “psd3.py,” “dimension_reduction_visualizations.py”).
• Additional utility scripts (like “echo”) and raw EEG data files for initial testing.
• Also, configuration and workflow files within .github and project-level files (like README.md and Roadmap) provide context on deployment and future design iterations.
a. Main Components to Include:
• Data Ingestion and Collection Layer:
• Data sources (EEG sensors, accelerometer data, possibly pose estimation inputs from external computer vision algorithms).
• DataCollection modules (including the GUI, data_collection.py, and file assets serving as inputs/ stimuli).
• File system components with a distinction between raw data storage (EEG and accelerometer CSVs) and supplementary data (subject info, procedure documents).
• Data Preprocessing and Visualization Layer:
• Scripts for cleaning and organizing data (data_visualization.py, modules/influx_data.py, and local_storage.py).
• Preprocessing pipelines that prepare data for modeling (filtering, normalization, segmentation).
• Model Development and Evaluation Layer:
• Preprocessing utilities in the ModelDevelopment directory.
• Model building components: Autoencoders for noise reduction; LSTM/Transformer-based predictive models; and potential CNN-based pose estimation integration.
• Real-time processing modules (realtime.py) responsible for making predictions.
• Evaluation and visualization of model outputs (dimension_reduction_visualizations.py, echo, etc.).
• External Dependencies and Integration Points:
• External BCI and pose estimation systems.
• Data augmentation pipelines and potential integration points for ensemble techniques.
• Deployment pipelines (workflow configs in .github/workflows).
b. Relationships and Interactions:
• Data flow from raw EEG collection to preprocessing, feeding into the Machine Learning models.
c. Architectural Patterns and Design Principles:
• Separation of concerns: data ingestion, preprocessing, and model training/prediction are isolated components.
• Hybrid pipeline design that allows modular testing (using separate directories and components).
• Use the layered architecture approach: a bottom layer (data collection), a middle layer (data processing/preprocessing) and a top layer (model development and real-time prediction).
d. Technologies and Frameworks:
• Though the file tree does not explicitly mention external libraries, note that deep learning frameworks (TensorFlow, PyTorch), signal processing tools, and visualization libraries are implicit components.
• Mention that the project utilizes a Convolutional Neural Network for pose estimation and Autoencoders, LSTM, Transformer models for EEG data analysis.
Capturing Essential Architectural Elements
• Detail every major component you identified in the file tree.
• Remember to incorporate the hybrid nature of the project: it’s not only collecting data but also actively running sophisticated real-time model predictions.