CRESP is a toolkit for standardizing and validating reproducibility of computational scientific experiments. This project aims to address reproducibility issues in computational science experiments and help researchers without programming expertise conduct computational research.
- Provides standardized configuration schemes for describing computational experiment environments and workflows
- Automatically records and validates outputs at each stage of the experiment
- Three validation levels:
strict: All output file hashes must match exactlystandard: Key results must match but allows differences in non-critical outputs; fixed random seeds but tolerates platform differencestolerant: Results are acceptable within specified ranges
- Lowers technical barriers for researchers to implement reproducible experiments
- Integrates with existing tools (like Pixi)
pip install crespcresp --help-
Configuration Management (core/config.py)
- YAML-based configuration schema
- Environment validation
- Path management
-
Hash Calculation (core/hashing.py)
- File hash computation
- Directory hash computation
- Hash comparison utilities
-
Random Seed Management (core/seed.py)
- Global seed control
- Per-stage seed management
- Framework-specific seed handling
-
Stage Management (core/stage.py)
- Stage definition and validation
- Stage dependency resolution
- Stage execution tracking
-
Validation Logic (core/validation.py)
- Multi-level validation strategies
- Output comparison
- Tolerance handling
- Package Initialization (init.py)
- Public Function Exposure
- API Documentation
- Basic Command Structure
- Subcommand Implementation
- init
- validate
- run
- report
-
Report Generation
- HTML reports
- Markdown reports
- Validation summaries
-
Tool Integration
- Pixi integration
- Git integration
- CI/CD support