ncar / esmlab-project

esmlab-project

Tools for a reproducible workflow.

esmlab-project's Issues

What is the objective of this repo?

A high-level objective is to enable "reproducible" science. In this vein, one might imagine being able to recreate a scientific workflow from end-to-end. Inputs might include obs, models, etc.; outputs include presentations, papers, and documentation of the algorithms used.

I recognize the following challenges in this context.

• Code for a particular project often has both general and specific components—all of it perhaps under active development. Need to document and enable best practices for co-development so as to leverage specific scientific-application driven development for general purposes.

• Ultimately a project might have many dead-ends. It might make sense to have a "scratch-pad" space; as components reach maturity, they could be moved over a divide into a "publication" directory.

• Scripts/notebooks generate figure files (pngs, pdfs, etc.); these are often given spur-of-the-moment names and end up being used in presentations, papers, etc. Tracking the provenance in such a setting is difficult. Figures generation could be wrapped in a class that enables provenance tracking; this might have a caption class that retains a description; could be easily harvested for web/markdown documentation of the project.

• environments need to be curated; it should be possible to re-invoke the specific state (i.e., package versions, etc.) of a general environment definition, for instance, to reproduce the project state at last execution. In my current workflow, it's incumbent on me (using conda) to activate an environment definition. Is it possible to ensure that a project is executed in the project environment?

How do we address these (and other) challenges? Do we create a "template repo"?