This module improves memory management in N2P2. More specifically, a new strategy to store symmetry function derivatives is implemented. In this way the memory footprint during training is drastically reduced.

Roadmap

• Create Boost tests for symmetry function calculation and training.
• Implement "element-combination-aware" storage for symmetry function derivatives.
• Provide example and check memory footprint before/after.
• Source code documentation.
• Complete module README.

Module verification checklist (for reviewers)

Checklist when the module is first submitted

• Have the relevant labels been added to the MR
• If submitted on someone elses behalf, has the software author been referenced (if they have a GitLab account)

Checklist when module is no longer "WIP"

• Is the module documentation sufficiently detailed?
• Is it mergeable? (i.e., there should be no merge conflicts)
• Are the build instructions sufficient - source code locations, build instructions, etc.? (If not the MR should be updated)
• Did it pass the tests that were described? (Are there unit/regression tests? Do they pass?)
• Are the tests sufficient?
• If the module introduces new functionality, is it tested? (Unit/regression tests?)
• Is the associated source code well formatted? (typos, line length, brackets,...it should be consistent with existing source)
• Is all new source code sufficiently documented? (functions, their arguments,...)
• Is there a description of any applications the module has? (This is a hard requirement for E-CAM PDRAs)

After Merging

• Make sure the module appears in a toctree
• Add a link to the final result on https://e-cam.readthedocs.io