Scripts Reference¶
The scripts/ directory contains automation scripts for building, testing, and managing BioML-bench environments and agents.
Build Scripts¶
build_base_env.sh¶
Builds the foundational Docker image (biomlbench-env) with biomedical libraries and dependencies.
Usage:
Features: - Installs Python 3.11 with conda - Includes biomedical libraries (RDKit, BioPython) - Configures ML frameworks (TensorFlow, PyTorch) - Sets up the grading server environment - Performs post-build validation tests
Requirements: - Docker installed and running - Internet connection for downloading dependencies - ~10GB disk space for the image
Example Output:
🧬 Building BioML-bench Base Environment
=======================================
📋 Pre-build checks...
✅ environment/Dockerfile
✅ environment/requirements.txt
🔨 Building biomlbench-env Docker image...
✅ Successfully built biomlbench-env image
🧪 Testing base image...
✅ Python is available
✅ BioML-bench is importable
✅ Biomedical and ML dependencies are available
✅ Agent conda environment is ready
🎉 Base environment build completed successfully!
build_agent.sh¶
Builds Docker images for individual agents.
Usage:
./scripts/build_agent.sh <agent-name>
# Examples
./scripts/build_agent.sh dummy
./scripts/build_agent.sh aide
Features:
- Builds agent-specific Docker images
- Inherits from biomlbench-env base image
- Configures agent-specific dependencies
- Sets up environment variables and entrypoints
Agent Directory Structure:
agents/<agent-name>/
├── Dockerfile # Agent-specific build instructions
├── config.yaml # Agent configuration
├── start.sh # Agent execution script
├── requirements.txt # Additional dependencies
└── src/ # Agent source code
Additional Features: - Multi-stage builds for smaller images - Build caching optimization - Security scanning integration - Performance benchmarking - Automated testing of built images
Testing Scripts¶
test_environment.sh¶
Comprehensive testing of the BioML-bench environment setup.
Usage: