fix(gfql): restore identity-stable _pl_nan_to_null (regression from #1731)#1733
Closed
lmeyerov wants to merge 2 commits into
Closed
fix(gfql): restore identity-stable _pl_nan_to_null (regression from #1731)#1733lmeyerov wants to merge 2 commits into
lmeyerov wants to merge 2 commits into
Conversation
…1731) #1726 (3809ecd) made Engine._pl_nan_to_null identity-stable: probe is_nan().any() per float column and return the SAME frame object when no NaN is present, because fill_nan(None) on a NaN-free column is a value-level no-op that still yields a fresh frame, and that identity churn defeats the #1658 CSR index's `source_ref is df` residency check (so a resident index is silently invalidated across a df_to_engine re-conversion on polars/polars-gpu, falling back to the O(E) scan). #1731's "final hardening" (c2138b0) accidentally reverted this to the unconditional df.with_columns([fill_nan(None) ...]), re-breaking identity for any Float32/Float64 frame. This restores the identity-stable probe (rebuild only the columns that carry a genuine NaN; rare LazyFrame path keeps the unconditional rewrite). Value-identical; only object identity changes. Validated on dgx (26.02-gfql-polars): 113 passed (restored TestPlNanToNull + full 4-engine test_index.py parity suite). Mechanism proof at the df_to_engine boundary — regressed: identity preserved False; fixed: True (NaN-present still converts, null_count 1). Pokec (all-int/str) was unaffected via the no-float-columns fast return; SNB/LDBC and graph-bench (float columns) regressed. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Y6dQEcjdazEnzuvuwf73ZL
…NaN case Review nit N1: the original tests used a single float column, so they did not exercise the "rebuild only the NaN-carrying columns" branch. Add a case with a clean float column + a NaN-carrying float column, asserting only the dirty column is converted (nulls introduced), the clean column is byte-identical, and column order + non-float columns are preserved. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Y6dQEcjdazEnzuvuwf73ZL
Contributor
Author
|
Closing as subsumed by #1735 (perf/gfql-pl-nan-clean-cache): #1735 contains this PR's identity-stable |
pull Bot
pushed a commit
to admariner/pygraphistry
that referenced
this pull request
Jul 20, 2026
… per-hop NaN re-scan on polars) Every hop() runs _coerce_input_formats -> _pl_nan_to_null, which scanned is_nan().any() over every float column of the (unchanged) resident edge frame ON EVERY CALL. So repeated seeded hops on a resident graph (seeded-Search / native-hop) grew O(E) with edge count on polars/polars-gpu while pandas stayed flat. _pl_nan_to_null now probes an eager polars frame for real NaN once, returns a clean frame UNCHANGED (restores the graphistry#1726 identity guard reverted by graphistry#1731 -> frame-identity caches like the graphistry#1658 index keep engaging), rewrites only columns that genuinely carry NaN (values identical to the old unconditional fill_nan), and caches the id of clean frames so later calls skip the probe. Recycle-safe via weakref.finalize (a rebound/new frame re-probes -> NaN->null semantics unchanged). Measured (dgx): indexed polars g.hop on 8M edges + one float column: 33.8ms (growing O(E)) -> 0.20ms FLAT (~140x, O(degree)), matching pandas. This is the per-call cost that made polars/polars-gpu seeded LDBC Search grow with scale and lose to Kuzu. Regression: 184 passed (full 4-engine test_index.py parity + test_engine_coercion) — the one failure (test_chain_dask_edges) is pre-existing on clean master (RAPIDS-image dask issue). New TestPlNanCleanCache: NaN-present cleaned, clean-frame cached same-object, distinct frames independent. Subsumes graphistry#1733 (identity-stable reland) and adds the clean-cache. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Y6dQEcjdazEnzuvuwf73ZL
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
Restores the identity-stable
Engine._pl_nan_to_nullfrom #1726 (3809ecd9), which #1731's "final hardening" (c2138b0d) accidentally reverted. The revert is live in current master and silently re-broke the #1658 seeded-index engagement on Polars / Polars-GPU for any graph carrying a float column.Root cause
_pl_nan_to_nullreturned a fresh polars frame whenever anyFloat32/Float64column existed (keyed on column presence, not actual NaN).fill_nan(None)on a NaN-free column is a value-level no-op but still yields a new object, and that identity churn defeats the #1658 CSR index'ssource_ref is dfresidency check — so a resident index is invalidated across adf_to_enginere-conversion and the query silently falls back to the O(E) scan.#1726 fixed this by probing
is_nan().any()per float column and returning the same object when no NaN is present.c2138b0drestored the unconditionaldf.with_columns([fill_nan(None) ...]), reintroducing the churn. This PR re-lands the identity-stable probe (rebuild only — and only the columns where — a NaN is genuinely present; the rare LazyFrame path keeps the unconditional rewrite).Impact
engine='polars'/'polars-gpu'(SNB/LDBC, graph-bench) — resident feat(gfql): physical adjacency indexes for O(degree) seeded traversal #1658 index invalidated across re-conversion → O(E) scan.if not float_cols: return dffast return).Validation (dgx-spark,
26.02-gfql-polarsimage, via safe_run)TestPlNanToNull(4 unit tests) + the full 4-enginetest_index.pyparity suite (pandas/cuDF/polars/polars-gpu).df_to_engineboundary: regressed →df_to_engine(polars→polars)identity preserved False; fixed → True, with NaN-present input still converting to null (null_count=1). Value-identical to prior behavior; only object identity changes.gfql_index_all+gfql_explainpath did not surface a difference by itself (it reuses the resident polars frame without re-conversion); the impact is confined to paths that re-crossdf_to_engine(the SNB native-hop lane, per GFQL: #1658 index does not engage on polars/polars-gpu direct hops (frame-identity churn) #1726's original measurement of 120/120-scan → 54-index).Change
graphistry/Engine.py— restore identity-stable_pl_nan_to_null(+ comment noting the GFQL: #1658 index does not engage on polars/polars-gpu direct hops (frame-identity churn) #1726→feat(gfql): add aggregate fast paths and final hardening #1731→re-land history).graphistry/tests/compute/test_engine_coercion.py— restoreTestPlNanToNullas a regression fence.CHANGELOG.md— Fixed entry.Regression fix only; no behavior change for NaN-carrying frames or non-float frames.
🤖 Generated with Claude Code