Conversation
lmeyerov
force-pushed
the
docs/review-skill-vectorization-cudf-checks
branch
from
243163f to
c597214
Compare
Adds a "Vectorization & engine compatibility (GFQL / row pipeline / compute)" subsection under "Pygraphistry-specific review checks" so reviewers consistently catch: * Per-row Python loops (`apply(axis=1)`, `iterrows`, `itertuples`, `loc[i, col] = ...` in a loop) on hot row paths — both a perf cliff on pandas and an outright break on cuDF. * cuDF compatibility regressions — pandas-only API patterns, GPU↔CPU round-trips that signal missing vectorization, signatures locked to `pd.DataFrame` instead of the engine-neutral `DataFrameT`. * When a code change requires paired cuDF coverage vs. when CPU-only is acceptable. Severity rubric and false-positive guards included so reviewers distinguish hot row paths from control-plane code and don't flag pre-existing repo debt as new findings. The pre-existing GPU-validation-evidence rule (line 172-174) keeps its current scope; this subsection adds the static-review-time checks that should fire BEFORE running GPU CI. Catalyst: pygraphistry #1279 (free-form intermediate MATCH admit) made the vectorization + cuDF assumptions concrete enough to codify.
lmeyerov
force-pushed
the
docs/review-skill-vectorization-cudf-checks
branch
from
c597214 to
b0865bc
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Adds three audit checks to
agents/skills/review/SKILL.md's "Pygraphistry-specific review checks" section so reviewers consistently catch GFQL/compute regressions:apply(axis=1),iterrows(),itertuples(), per-row Python loops, scalar-in-loop access on hot row paths.df[col]=v,df.loc[mask,c]=v, cell-wise mutation in loops,inplace=True,del df[col],df.append(...), mutating-then-returning-input. Inline pure alternatives (df.assign,df.where,df.drop, engine-polymorphicdf_concat(engine), etc.)..to_pandas()round-trips,is pd.NAcomparisons, dtype string matches, signatures locked topd.DataFrameinstead ofDataFrameT.Plus: a list of engine-polymorphic helpers reviewers should recommend (
df_cons,df_concat,df_to_engine,safe_merge,resolve_engine,template_df_cons,DataFrameT); paired-cuDF-coverage rule; flag-wording template; false-positive guard distinguishing hot-row paths from control-plane code.Why now
The skill currently has one GPU-related line ("if the PR touches GPU code, run GPU CI on dgx-spark") that fires AFTER a code change ships. There's nothing pre-CI guiding a reviewer to flag, e.g., a new
df.apply(fn, axis=1)on the row pipeline (cuDF-incompatible), a newdf[col] = vmutation pattern, aninplace=True, or a helper signature locking the DataFrame param topd.DataFrame.Pygraphistry #1279 (free-form intermediate MATCH admit, in review) made these assumptions concrete enough to codify. This PR is independent of #1279.
What's in / out
In: a
#### Vectorization & engine compatibility (GFQL / row pipeline / compute)subsection (~53 LOC under "Pygraphistry-specific review checks"). Three terse severity tables, engine-polymorphic-helpers list, paired-cuDF rule, flag wording, FP guard.Out: GPU-validation-evidence rule (line 172-174) unchanged. No changes to other skill phases.
Test plan
Refs