How egordm/mgtcom works
Main Components
a) Core Framework Components:
• Data Processing Layer
• Model Variants Layer
• Embedding Generation Layer
• Community Detection Layer
• Evaluation Layer
b) Supporting Components:
• Dataset Management System
• Model Training Pipeline
• Baseline Models Integration
• Evaluation Metrics System
Detailed Architecture Breakdown
a) Data Layer:
• Data preprocessing pipeline (datasets/transforms/)
• Temporal and static data handling (datasets/temporal/ and datasets/static/)
b) Model Architecture:
• Create three parallel branches for model variants:
• MGTCOM (Combined temporal and topological)
• MGTCOM Topo (Topological only)
• MGTCOM Tempo (Temporal only)
c) Core Processing Pipeline:
• Data Sampling (ml/ml/data/samplers/)
• Feature Extraction (ml/ml/layers/)
• Embedding Generation (ml/ml/models/)
• Community Detection (ml/ml/algo/clustering/)
• Model Training (ml/ml/executors/)
Integration Points
• Connections between baseline models (GraphSAGE, Node2Vec, etc.)
• Evaluation pipeline (ml/ml/evaluation/)
• Callback system integration (ml/ml/callbacks/)
Specific Design Elements
Use the following visual elements:
• Blue rectangles: Core components
• Green rectangles: Data processing components
• Orange rectangles: Model variants
• Purple rectangles: Evaluation components
Important Details
• Data flow directions between components
• Temporal aspect handling
• Multimodal graph processing
• Community detection algorithms
• Integration points with baseline models
• Training and evaluation loops
• Configuration management (from ml/ml/utils/config.py)
Technical Specifications
• Python dependencies
• CUDA requirements
• Memory requirements
• Supported input formats
• Output formats