Dataset insertion benchmark

[hmaarrfk]

xref: #7221

• Closes #xxxx
• Tests added
• User visible changes (including notable bug fixes) are documented in whats-new.rst
• New functions/methods are listed in api.rst

[hmaarrfk]

ok naming is always hard. I tried to pick a good name.

[max-sixty]

Excellent!

[hmaarrfk]

TBH, i was going to expand this to other in-memory operations. Replace a variable? Drop a variable?

[Illviljan]

I thought the first version used was relevant as well:

ds = xr.Dataset(coords=coords)
for v in names:
    ds[v] = ("time", value)

And this fast one from #7224 (comment) would also be nice to benchmark:

data_vars = {v: ("time", value) for v in names}
ds = xr.Dataset(data_vars=data_vars, coords=coords)

[hmaarrfk]

The first case is captured by cause i am suggesting we time insertion for 0, 10, 100, 1000 existing variable. One can interpolate the bahvior.

I do believe that bencharmming all at once creation (the second case) is also interesting

[dcherian]

I renamed it thinking that we were really timing the merge operations. This is also true for replacing, but not true for dropping.

Maybe we add dropping to a new dataset.py. This is all quite minor though! Shall we merge?

[hmaarrfk]

thinking that we were really timing the merge operations

It depends if you want to think about from the point of view of "xarray developer" or "xarray user".

As an "xarray user", i'm trying to "insert a totally new variable, no dims, no coords". I don't really care what is happening under the hood. This is what I originally had in mind. This is h

My goal with the original name was to test operations on in-memory xarrays (as opposed to those backed by a file, which seem to be done in dataset_io). I don't know if I have time to "add other benchmarks" at the moment. This seems good.

The name is the only thing that will survive. so if you are are ok with it.

I ran the benchmarks against main, #7221, and #7222. The benchmark seems to show the effect locally even with a single insertion. I've thus simplified the benchmark to do a single insertion, and not 5.

main

[ 50.00%] ··· merge.DatasetAddVariable.time_variable_insertion   ok
[ 50.00%] ··· ======== ==========
               param1
              -------- ----------
                 0      636±0μs
                 10     511±0μs
                100     5.92±0ms
                1000    41.0±0ms
              ======== ==========

#7221

[ 50.00%] ··· merge.DatasetAddVariable.time_variable_insertion   ok
[ 50.00%] ··· ======== ==========
               param1
              -------- ----------
                 0      609±0μs
                 10     1.32±0ms
                100     4.01±0ms
                1000    8.65±0ms
              ======== ==========

#7222

[ 50.00%] ··· merge.DatasetAddVariable.time_variable_insertion   ok
[ 50.00%] ··· ======== ==========
               param1
              -------- ----------
                 0      578±0μs
                 10     1.03±0ms
                100     2.74±0ms
                1000    5.37±0ms
              ======== ==========

[hmaarrfk]

Is there a way rename param1? ok i renamed it.

[hmaarrfk]

$ asv run -E existing --quick --bench merge.DatasetAddVariable.time_variable_insertion
· Discovering benchmarks
· Running 1 total benchmarks (1 commits * 1 environments * 1 benchmarks)
[  0.00%] ·· Benchmarking existing-py_home_mark_mambaforge_envs_mcam_dev_bin_python
[ 50.00%] ··· ...e.DatasetAddVariable.time_variable_insertion                 ok
[ 50.00%] ··· =================== ==========
               existing_elements
              ------------------- ----------
                       0           668±0μs
                       10          1.17±0ms
                      100          2.22±0ms
                      1000         5.38±0ms
              =================== ==========

[dcherian]

Thanks @hmaarrfk Great PR!