pandas2
pandas2 copied to clipboard
DEV: Publishing benchmarks vs. trunk?
We'll want to develop a microbenchmark suite to compare pandas 2.0 perf versus 1.x, especially in microbenchmarks. What's the right tool for this? asv, vbench?
IMO certainly not vbench (unless someone wants to put time in developing, but even then). The benchmark suite already is ported over to asv, and I have a PR to clean them up (https://github.com/pandas-dev/pandas/pull/14099, once that is finished, I would remove the vbench ones, we don't use them anymore). But I have less of an idea how reliable it is for microbenchmarks (I think it is based on timeit, and I think there are arguments to be made that this is not really reliable for microbenchmarks)
I was thinking that ASV is the best choice -- as long as we can easily compare perf vs. a snapshot of master (e.g. we can use v0.19.0 as a baseline) so we can see how we're doing.