How echkidieu1102/ca-gcn works
Main Components
• Data Layer: Represent the datasets organized under the “data” directory, which include multiple knowledge graph datasets with associated entity/relation mappings, training/validation/test splits, and even historical sequence information (especially for ICEWS14).
• Data Preprocessing / History Extraction: Highlight the role of scripts such as “get_history_record.py”, which likely parse and prepare historical sequence data from the datasets before training/inference.
• Model Definition and Core Engine: Capture key modules in the “rgcn” directory (which includes files like knowledge_graph.py, layers.py, and model.py) and the complementary implementation files in the “src” folder (main.py, model.py, rrgcn.py). These collectively build and define the CA-GCN model.
• Training / Inference Pipeline: Mark out the entry point “src/main.py” as the orchestrator that leverages the model definitions, processes input data, triggers training routines, and eventually carries out evaluation.
• Output and Logging: Include “models” (where trained models are saved) and “logs” (which contain log files for different datasets such as ICEWS18 and WIKI), representing the system’s output and monitoring components.
• From the preprocessing component, indicate that prepared data is fed into the training/inference pipeline (via src/main.py) which interacts with the model component (the GCN modules found in “rgcn” and “src”).
Architectural Patterns
• Modularity: The separation of concerns is clear—data management (file structure under “data”), model definition (inside “rgcn” and “src”), and pipeline management (main.py, logs, and models).
• The design follows a typical research/experimental ML architecture with a distinct training loop, model evaluation, and logging processes.
• There is also evidence of temporal or dynamic information handling given the extensive set of historical sequence files.
Technologies and Frameworks (as relevant)
• Although not explicitly specified, you might note that the project is likely built with Python and possibly leverages PyTorch or TensorFlow for neural network operations (this is common in such research code).
• The use of npz files suggests that NumPy is used extensively for data storage or preprocessed arrays.