Main talking points
Challenges in Reproducible Science:
- Difficulties in iterating on different model types and scaling up for various configurations.
- The need for tools that abstract away complexities and promote rapid prototyping.
- The challenge of deciding how long we want to reproduce old code versus finding modern libraries that achieve the same results, given the rapid pace of technological change.
- The idea of tracking future libraries that can run older code, especially from the perspective of the “map of open source science.”
Tools and Solutions:
- Introduction of Covalence, a workflow orchestration platform for quantum and classical computing.
- Emphasis on Covalence’s ease of use and relevance to reproducible science.
- Discussion on the importance of data integration for machine learning models, especially in compliance-heavy industries.
- Tools in the Python ecosystem that help record versions used in pipelines.
Best Practices in Code Development:
- The significance of adhering to best practices for long-lasting and stable code pipelines.
- Challenges with versioning libraries in Python and the importance of reproducibility.
- Experience with using Docker for pipelines and issues with Conda channels.
- The importance of replication and the challenges associated with it.
- The significance of credit for researchers and software developers in open-source research software.