How aidanallchin/cse6242-patentproject works
Project type and purpose
• This is a data‐processing and analysis pipeline (not a web app). It ingests patent and economic data, cleans/enriches it (including a large LLM for geocoding), generates county‐level features, trains a predictive model (LightGBM), and outputs visual overlays for Tableau.
File structure
• Top‐level script:
• main.py and __init__.py orchestrate setup, directory creation, downloads, and menu‐driven pipeline steps.
• src/data_cleaning: all modules that ingest raw USPTO/BEA/Census files, clean patents, assign lat/long, map to FIPS.
• src/modeling_overlays: generate_predictors.py and model.py for feature assembly and training the Innovation Hub Predictor Model.
• src/other: helpers.py and logging.py for shared utilities.
• Config/dependencies: requirements.txt; PatentPulse_Stats.twb is the Tableau workbook; images/model_flow_diagram.png is a reference flowchart.
Key insights from README
• Sequential ETL → cleaning → feature generation → modeling → visualization.
• External data sources: USPTO S3 endpoints, BEA ZIPs, Census GeoJSON.
• Optional manual downloads to save time (three intermediate files).
• Uses a 32B‐parameter LLM for geolocation corrections.
• LightGBM with DART boosting for prediction, output visualized on a map in Tableau.
Final recommendations
• Be exhaustive: represent every top‐level module/directory as a component.
• Auxiliary utilities (helpers, logging) as side‐components.