It may be useful, if one could directly modify final results after convergence. I do not know whether this something that people would like to have.
I could think of a lot of scenarios where this may come in handy.
If I am not mistaken, the code could be changed here:
https://github.com/PrivateAIM/python-sdk-patterns/blob/28d8df32215f358d080925c9afc684376a793931/flame/star/star_model.py#L77C1-L84C69
As for examples, I would think of cases where one is interested in getting further information that is not directly used in every round of analysis/aggregation, e.g.
- transformation of final results, e.g. better format, additional transformations etc.
- extending final results (e.g. detailed info of convergence)
- reducing results by compressing, thinning, summarizing
Currently, a workaround is to communicate in each round an additional convergence variable and make an extra 'pseudo round' of analysis-aggregation with transformed results. See for instance here where I modify the final results to add common summary statistics.
If you think, that this is not needed, you can ofc close the issue :)
It may be useful, if one could directly modify final results after convergence. I do not know whether this something that people would like to have.
I could think of a lot of scenarios where this may come in handy.
If I am not mistaken, the code could be changed here:
https://github.com/PrivateAIM/python-sdk-patterns/blob/28d8df32215f358d080925c9afc684376a793931/flame/star/star_model.py#L77C1-L84C69
As for examples, I would think of cases where one is interested in getting further information that is not directly used in every round of analysis/aggregation, e.g.
Currently, a workaround is to communicate in each round an additional convergence variable and make an extra 'pseudo round' of analysis-aggregation with transformed results. See for instance here where I modify the final results to add common summary statistics.
If you think, that this is not needed, you can ofc close the issue :)