How frankl1/sast works
Main Components
• Core Algorithm Module:
• Centralized SAST implementation contained in the "sast" directory (e.g., sast.py for the core SAST algorithm, stck.py for STC variants).
• Utility functions (utils.py) that support common operations.
• Experiment Scripts:
• Consolidate all experiment files under the "experiment-scripts" folder. Each script (such as experiment-sast-approx.py, experiment-sast-ensemble-full.py, etc.) demonstrates different aspects or variations of the shapelet-based approaches.
• That these scripts include benchmarking experiments (accuracy comparisons, scalability tests) that feed data into the classifiers and produce CSV results.
• Data & Result Handling:
• "results" directory as the sink where outcomes of experiments are stored in CSV formats.
• Notebook & Documentation:
• The “sast.ipynb” Jupyter notebook acts as an interface or tutorial demonstrating usage.
• The README provides contextual explanation, sample code, and dependency requirements.
Relationships and Interactions
• That these modules rely on utility functions (e.g., from utils.py) for data processing and common computational tasks.
• Specify the flow: Input Data → Preprocessing → Core Algorithm (SAST/STC) → Classifier Fitting/Predicting → Results Generation → Visualization (images, CSV output).
Architectural Patterns & Design Principles
• Modularity: separate directories for core algorithms versus experimental scripts and result visualization.
• An experimental/research workflow pattern: core module → experiment orchestration → data collection and visualization.
• Lay out dependency and integration relationships with external libraries (numpy, numba, scikit-learn, sktime) which support numerical computations and model building.
Technologies and Frameworks
• Usage of Python and core numerical libraries (numpy, numba) and ML tools (scikit-learn).
• Optionally mention the integration with the Aeon toolkit as per the README links, indicating that SAST has been incorporated or referenced via that broader ecosystem.