jiayuasu opened a new issue, #133: URL: https://github.com/apache/sedona-spatialbench/issues/133
Follow-up to #131, which adds the geography query suite. `notebooks/queries.ipynb` ships **with** executed outputs — every query cell has a result table, and those tables are what appear in `docs/queries.md`. `notebooks/geography-queries.ipynb` was committed **without** outputs, because executing it today would put misleading results into the docs. ## Why it was deferred Two SedonaDB defects, both filed and both engine-side: - **apache/sedona-db#1084** — a geography scalar function over the output of a geography spatial join fails with `Array input must not be empty`. Q9 raises this, so its cell would carry an error traceback. - **apache/sedona-db#1085** — one zone polygon of 156 095 (`z_zonekey` 150480) returns a negative geography `ST_Area` and matches points anywhere on Earth. Q4, Q6, Q10 and Q11 are all distorted by it: at SF1, Q6 would display `Shendi | 5999980` — essentially every trip in the dataset, for a county nowhere near the query area — and Q11 would report 6 290 682 cross-zone trips against the planar suite's 176 391. Committing those outputs would document engine bugs as if they were the queries' results, and the page deliberately carries no engine-defect discussion. Executing only the sound subset was considered and rejected: a notebook where some cells have outputs and others do not reads as broken rather than deliberate. ## What to do Once #1084 and #1085 are released: 1. Install the docs environment (`docs/requirements.txt`, which brings `apache-sedona[db]`) and generate or download SF1 data. 2. Run `notebooks/geography-queries.ipynb` top to bottom. Q12 is a markdown cell, not a code cell, so it will not error — it needs a geography `ST_KNN` (apache/sedona-db#1086, apache/sedona#3178). 3. Regenerate the page: `python -m nbconvert --to markdown notebooks/geography-queries.ipynb`, then re-prepend the YAML front matter and ASF license header that `docs/geography-queries.md` carries (the notebook itself is RAT-excluded via `*.ipynb`; the page is not, unlike `docs/queries.md`). 4. Run `pre-commit run --files ...` — the end-of-file fixer adjusts nbconvert's trailing newlines. 5. Mirror the new output blocks into `docs/geography-queries.zh.md`, which is maintained by hand from the English page. ## Sanity checks worth keeping At SF1 the geography results were verified against the committed geometry answers: Q2 matches exactly (541), Q3 matches its row count (22), Q4 reproduces the geometry answer row for row (258 zones, same counts, same order) once the #1085 polygon is excluded, and Q9's self-join finds the same 37 building pairs. Those are good regression anchors when the outputs are finally committed. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
