Add Pprof Profiling to Existing Benchmarks

[oojo12]

Resolves #272.

By the way running the benchmarks in the background and will update #228 once if this is approved.

[codecov-commenter]

Codecov Report

Base: 38.75% // Head: 38.68% // Decreases project coverage by -0.06% ⚠️

Coverage data is based on head (055eb24) compared to base (c301cee).
Patch has no changes to coverable lines.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #273      +/-   ##
==========================================
- Coverage   38.75%   38.68%   -0.07%     
==========================================
  Files          93       93              
  Lines        6087     6087              
==========================================
- Hits         2359     2355       -4     
- Misses       3728     3732       +4     

Impacted Files	Coverage Δ
...rithms/linfa-trees/src/decision_trees/algorithm.rs	37.94% <0.00%> (-1.79%)	⬇️
algorithms/linfa-nn/tests/nn.rs	75.60% <0.00%> (ø)

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[oojo12]

The windows failures are because pprof won't compile correctly there. I had to use wsl to compile. Is there something like bench_dependencies for cargo so we can resolve this? Tried googling but didn't find anything

[oojo12]

Actually just found docs

[YuhanLiin]

Do the benchmarks also output a text profile report?

[oojo12]

Not sure I can look into it later in the week. Example flamegraph.svg's from running this are attached. I'll add the relevant one to #262. Let me know if you prefer the proto format, to me it seems more user friendly.

criterion.zip

[YuhanLiin]

The zip file only contains the flamegraphs. I guess the profiler doesn't output any extra text right? Also, Criterion still outputs graphs just like before right?

As for the format, I still prefer flamegraphs since that's what I'm used to, but ideally I'd like to have both flamegraphs and the proto format.

[YuhanLiin]

As part of this PR, can you also update the Linfa benchmark standards? We should do it once the profiling code is finalized.

[oojo12]

The zip file only contains the flamegraphs. I guess the profiler doesn't output any extra text right? Also, Criterion still outputs graphs just like before right?

As for the format, I still prefer flamegraphs since that's what I'm used to, but ideally I'd like to have both flamegraphs and the proto format.

It is one or the other you either output flamegraphs or use criterion as the measurement method. See here for documentation evidence. As for the format I believe it is either flamegraph format or proto format. I don't think you can do both without running the benchmark twice. Not a problem to add just not sure if the value add is worth it.

[oojo12]

As part of this PR, can you also update the Linfa benchmark standards? We should do it once the profiling code is finalized.

yep once we agree on this PR being ready for merge I plan to.