Update OneShiftOnly doc#5155
Conversation
| If optimization gets stuck with proposed parameter sets being constantly rejected, decrease ``minwalkers``. | ||
| When the energy gradually converges, keeping a large ``minwalkers`` is necessary to prevent accepting risky parameter sets. | ||
| Continuing optimization with more parameters from a partially converged wavefunction should also use large ``minwalkers``, | ||
| for example adding three-body Jastrow factor to converged one-body and two-body Jastrow factors. |
There was a problem hiding this comment.
It sounds like this process is expected to take several job submissions, Is it actually made clear elsewhere in the manual how to do this?
There was a problem hiding this comment.
Peter: We need more examples. Another strategy is also to freeze/unfreeze parameters.
This updated text is a good improvement. It needs a little smoothing for language and I think specific examples of small and large values for minwalkers in the text would help with readability. One also has to search to find the acceptable range of values for minwalkers and what it is.
There was a problem hiding this comment.
@PDoakORNL It can be done with several jobs, it can also be done within a single job using https://qmcpack.readthedocs.io/en/develop/methods.html#variational-parameter-selection with a bit waste of computation evaluating zeros.
To reach control free optimization, we need to minimize the set of controlling knobs, understand how they behave and then develop a smart scheme. As @PDoakORNL mentioned, examples are important and I also think some FAQs are needed.
There was a problem hiding this comment.
So this seems like an improvement for sure. Unless you're planning to write more now I think it could just go in.
prckent
left a comment
prckent
left a comment
There was a problem hiding this comment.
Updated. I'll start the CI and merge process but please check for typos.
|
Test this please |
3 participants
Proposed changes
Clarify how to choose and adjust
minwalkersWhat type(s) of changes does this code introduce?
Does this introduce a breaking change?
What systems has this change been tested on?
epyc-server
Checklist