How vcovo/enthalpy-prediction works
Main Components
a. Data Layer:
• It to a data processing component that generates the processed dataset
• Specialized datasets (octene and nonyne isomers) as separate data stores
b. Processing Pipeline:
• Feature engineering/selection flow
c. Model Training Layer:
• Split into two parallel tracks for SVR and ANN
• Grid search components for both models
• Model persistence (saving) functionality
d. Evaluation & Analysis Layer:
• Error estimation components for both models
• Sensitivity analysis module
• Results storage for grid search and analysis
Component Relationships
Raw Data → Data Processing → Processed Data
• ↓
• Model Training (Parallel)
• / \
• SVR ANN
• / \
• Grid Search Grid Search
• /
• /
• Error Estimation & Analysis
• ↓
• Inference/Prediction Module
Technical Details
• Input formats: Excel (.xlsx) and CSV files
• Model storage formats: .sav for SVR, .h5 for ANN
• Key hyperparameters for both models (as noted in README)
• Python environment dependencies
Additional Elements
• Version information for critical dependencies
• Validation/testing workflow with 10-fold cross-validation
• Where sensitivity analysis fits in the pipeline
Key Workflows to Highlight
a. Training Workflow:
Data → Processing → Grid Search → Model Selection → Final Model
b. Inference Workflow:
Input Data → Processed Features → Model Inference → Prediction
Remember to include a note about the project's academic nature and its connection to the published paper, as this context is important for understanding the design choices.
This system design should effectively communicate both the research nature of the project and its practical implementation details to other engineers who might need to work with or extend the codebase.