Is there a way to benchmark the inspector?

[reikdas]

It seems like the benchmarking is tightly coupled with the execution. Is there a way to benchmark the inspector (the row grouping, the frequency based mining, etc) with a minimal diff?

[cheshmi]

The inspector is naively written because it does not matter much for pruned neural network. We did not optimize it after. I don't think we have the inspector timing in our logs now. I looked into the code, but the code has changed from the version that I worked on.