Build an iterative phonon flow #306
Conversation
|
Disclaimer: This is a small programming project that I might work on for fun in the next weeks. |
JaGeo
changed the title
…ex into iterative_phonon_flow.py
…ex into iterative_phonon_flow.py
…ex into iterative_phonon_flow.py
|
@QuantumChemist could you please review this PR? The code is working for me but along the way I also fixed other problems (glue.xml, test.ext.xyz full with additional data, removal of different displacements for the benchmark) that I only partially was able to test. I haven't tested the glue.xml part and there was no test in our current tests for that part. |
| if suffix == "without_regularization": | ||
| suffix = "without_reg" | ||
| if re.match(r"job_\d{4}-\d{2}-\d{2}-\d{2}-\d{2}-\d{2}-\d{6}-\d{5}", suffix): | ||
| if suffix not in ["phonon", "rattled"]: |
There was a problem hiding this comment.
Old version only works for jobflow but not jobflow remote
There was a problem hiding this comment.
also my mistake for not thinking about the jobflow remote structure of jobs when implementing this.
Co-authored-by: J. George <[email protected]>
we have one unit test for the MLIPFitMaker using glue.xml and I have extended it a bit to verify that the fit keeps working. |
…l/autoplex into iterative_phonon_flow.py
src/autoplex/auto/phonons/jobs.py
Outdated
| def get_output( | ||
| metrics: list, | ||
| benchmark_structures: list[Structure] | None = None, | ||
| benchmark_mp_ids: list[str] | None = None, | ||
| dft_references: list[PhononBSDOSDoc] | None = None, | ||
| pre_xyz_files: list[str] | None = None, | ||
| pre_database_dir: str | None = None, | ||
| fit_kwargs_list: list | None = None, | ||
| ): | ||
| """ | ||
| Job to collect all output infos for potential restarts. |
There was a problem hiding this comment.
the only thing I see code-wise is that this function name and its description coule be more descriptive. I needed to look at the unit test to fully get it 😃
There was a problem hiding this comment.
Otheriwse this PR looks good to me. :D
There was a problem hiding this comment.
Feel free to add more description 😀
Thanks. Yes, as a general feedback on the tests: we feel we need more asserts. Currently, we have many run-through tests. Maybe, you can open this as a new issue and identify tests where this is the case. |
Very first draft for an iterative phonon flow. There are still many, many things to solve before this can work.
make sure that a new random seed is used for each step (make sure the order of the settings has been considered in the seed computation)
reuse the previous database
make sure that the phonon benchmark runs are reused as well
expose all relevant outputs to allow for an iterative procedure
random_seed + len(workflow_maker.volume_custom_scale_factors)
fix previous unit tests
fix reading mace models
improve distort_type_1
make sure random seed creation is done correctly when very similar structures are provided in "structures" - currently, this is not correctly considered (needs to be done in iterative flow as well)
fix write_benchmark_metrics
document that only 0.01 displacement is used in the benchmark
add a warning if single displaced supercells are activated and run after iteration 0 (via postinit method)
do some testing on the cluster to confirm the options and capabilities
additional data ends up in text.extxyz. While this isn't directly used for fitting and plotting, this should not happen as users will not be able to understand where this data comes from.
make rmse computation more robust (add a try except block)
make sure all relevant data is part of the datastore (failures in
do_iterative_*)These points will be moved into a subsequent issue.