This repository is organized around a small shared core plus task-specific entry points.
astra_config.pyLoads.env, resolves dataset and cache paths, and centralizes provider-specific API settings.model_clients.pyWraps OpenAI-compatible APIs and local generation servers in a shared interface.
table2tree.pyTable-to-tree conversion logic.treeqa.pyTree-path retrieval, answer generation, and symbolic reasoning.tableqa.pyEnd-to-end batch pipeline and dataset loading.
evaluate.pyEvaluates prediction files.quality_evaluate/Measures tree quality.Batch_evaluate/Runs repeated generation for stability analysis.baseline/Direct and tree-direct baselines.demo/Interactive FastAPI + React application.
- Local paths now resolve through environment variables instead of hardcoded author machine paths.
- Shared model logic moved to
model_clients.pyto reduce duplicated provider logic. - Shared dataset and embedding configuration moved to
astra_config.py. - README and module docs now point to the current repository structure instead of removed files.