Task Development¶
Guide for creating new biomedical benchmark tasks in BioML-bench.
Task Structure¶
Each task requires:
config.yaml
- Task configuration and metadataprepare.py
- Data preparation logicgrade.py
- Evaluation functiondescription.md
- Task description for agents
Quick Start¶
-
Create task directory:
-
Add configuration file
- Implement preparation logic
- Define evaluation metrics
- Test with dummy agent
See Adding Tasks Guide for detailed instructions.