How minoru938/cspml works
Main Components
• Data Management & Storage:
• Data files and datasets are stored under both the top-level “data_set” directory and within “CSPML_latest_codes/data_set”. They include candidate crystals, structural data, pre-trained model files, and other arrays (e.g., .npy, .pd.xz).
• These datasets provide the material’s input for both training and predictive modeling.
• Model Training & Prediction Modules:
• Core algorithms are implemented in CSPML.py and complemented by additional scripts and notebooks (e.g., CSPML_training.ipynb, CSPML_Structure_Prediction.ipynb) that compare training outcomes.
• Pre-trained models (embedded as part of the repository) and the machine learning classification engine (binary neural network model) are key here.
• Updated/Latest Model Versions:
• The “CSPML_latest_codes” folder contains newer training codes addressing bias and newer TensorFlow versions, along with an updated training approach as described for the recent publication.
• User Interface & Tutorials:
• The tutorial.ipynb and various Jupyter notebooks serve as both documentation and interactive demonstrations.
• External Dependencies & Integration:
• The project relies on external libraries (pandas, numpy, tensorflow, pymatgen, xenonpy, torch, and optionally, matminer). Their interactions with data preprocessing, feature extraction (e.g., structure fingerprints), and model predictions should be represented.
Relationships and Interactions
• Data flows from the data storage layers (both “data_set” directories) into preprocessing and then into the machine learning pipeline.
• The updated “CSPML_latest_codes” plays a dual role: it mirrors the original module architecture but with enhanced training controls and additional processing pipelines.
• External libraries and dependencies should be depicted as side components that feed into the core components, emphasizing their roles (e.g., TensorFlow powering the training and "xenonpy" enhancing feature extraction).
Architectural Patterns and Design Principles
• Modular Design: Separate directories for datasets, core code, updated versions, and notebooks indicate that each major functionality is self-contained yet interconnected.
• Pipeline & Workflow Pattern: The system essentially works as a pipeline, where data is ingested, processed, used in model training, and finally served for prediction.
• Versioning for Research & Production: There is a clear distinction between the standard CSPML and the updated versions (contained in "CSPML_latest_codes"), emphasizing iterative improvements on the methodology.