Barycentric subdivision#13
Conversation
… on data's simplicial complexes. Then, another transformation can get the updated triangulation, to avoid overhead with previous models.
danielbinschmid
left a comment
danielbinschmid
left a comment
There was a problem hiding this comment.
Looks very good. Just some minor questions
|
|
||
| # add benchmarking results | ||
| results.add(data=out[0], config=config) | ||
| results.add(data=out[0][0], config=config) |
There was a problem hiding this comment.
does this still work for every configuration? Why is the result array now two-dimensional?
There was a problem hiding this comment.
Just saw the edit in run_experiment.py. I see now that we add a result for every barycentric subdivision. Shouldn't the line be something like results.add(data=out[idx][0], config=config) where idx is the index of the barycentric subdivision then?
There was a problem hiding this comment.
Yes! True! Good catch!
There was a problem hiding this comment.
Great, it looks better now in 25ff2a0
One more thought:
How is it intended to evaluate the results of the barycentric subdivisions? I think we need to add information about that an added benchmark corresponds to its relevant subdivision configuration in the ResultCollection.
Since the subdivision configuration it is not included in the ConfigExperimentRun, I think it can be done by adding another argument to the add method in the ResultCollection class (needs to also be handled in the save_result method).
|
We also need to merge the new commits in main to this branch |
2 participants
The pull requests allows one to test the trained neural networks on the barycentric subdivisions of the original test dataset specifying a maximum number of barycentric subdivisions to perform.