DOC: use constants in performance-comparisons.ipynb

[raybellwaves]

Description

I've simplified the performance comparisons notebook by setting constants which can be adjusted at the top of each section e.g. num_rows. This makes it easier for anyone running this to adjust the value and hopefully not encounter memory values. It can also help with testing these benchmarks on dataframes of various lengths. I've stripped the output as I was working on a A10G and I couldn't run with the current num_rows value. I also didn't want to commit the results which may differ compared to the H100 which is used currently and I would rather the results be committed by the RAPIDS team. I can confirm the notebook runs end-to-end (you can see my version here: https://github.com/raybellwaves/cudf-performance-comparisons/blob/main/performance-comparisons.ipynb with smaller num_rows and smaller timeit_number on a A10G (EC2 machine))

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[bdice]

@raybellwaves Thanks for the PR! I skimmed it and the core ideas seem helpful.

@galipremsagar Would you be able to take a look at this and incorporate the proposed changes for next time we run benchmarks?

[galipremsagar]

I'll run these on H100's and update this PR

[galipremsagar]

/okay to test

[bdice]

Can we fix (or hide) the warning here?

DataFrameGroupBy.apply operated on the grouping columns. This behavior is deprecated, and in a future version of pandas the grouping columns will be excluded from the operation. Either pass `include_groups=False` to exclude the groupings or explicitly select the grouping columns after groupby to silence this warning.
  lambda df: df.groupby(["key"], group_keys=False).apply(custom_formula_udf)

Also can we run on a system that isn't doing background work? I see some references to tritonserver consuming GPU memory (probably on other devices?) in the nvidia-smi output, which makes it hard to know if it affected performance negatively.

[mroeschke]

Nice changes. When printing out the cudf version, could you also include the pandas version as well?

[mroeschke]

My comment is non blocking so approving

[galipremsagar]

/okay to test

[galipremsagar]

/merge