Spaces:
Running
Running
Commit History
-
Trust ticks.parquet `t` and `event_seconds`
Upstream fixed the time bases: ticks `t` is monotonic video time and `event_seconds` is already on the POV video timeline. So: - world.ts uses `r.t` directly; drop the tick-derived recompute and the TICK_RATE constant. - map-results.ts drops the ×2 multiplier for `event_seconds`. - Recipes drop `event_seconds * 2.0` and the pov_rounds join that was only there for path lookups we don't need.1632a07
-
Minimap: time frames from tick column (t column is non-monotonic across pauses)
16c253d -
Minimap: fix ticks.parquet column names (x/y/z/yaw/armor_value/team_num)
6c84e50 -
Minimap: read ticks.parquet instead of world.preview.jsonl
437403c -
Point viewer at opencs2_dataset layout
Verified signature10d4db6
-
Hub API: read compacted web-index parquets
Home page: 454 reqs (1 tree + 226 manifest + 227 rounds shards) → 2 reqs (index/web/maps.parquet + index/web/rounds.parquet). Match page: 10 per-player chunks-preview shards → 1 per-(match,map) chunks-preview.parquet. Drops the now-unused listTree helper and PLAYER_COUNT constant.3f5bd49
-
Nerfies-style home, paper-style match header, prettier setup, dataset README
- Redesign home page: centered title, author + links, abstract, motivation cards, BibTeX - Slim down match-page header to a minimal back-link + theme toggle - Replace ternary class injections with data-attributes (score, match-info, view mode) - Drop unused shadcn surface (accordion, alert-dialog, breadcrumb, ...) and pnpm artifacts - Expand README with usage, schema, dev/build, viewer internals, BibTeX - Improve <svelte:head> meta tags and robots.txt - Add Prettier with svelte + tailwindcss plugins; format the entire codebase95e3d2a -
Add hyparquet-compressors so ZSTD parquets actually parse
The dataset's parquets are ZSTD-compressed and hyparquet's core only ships support for the trivially-decodable codecs. Pulled in `hyparquet-compressors` (brotli/snappy/lz4/zstd/...) and pass its `compressors` map into `parquetReadObjects`. Adds ~100KB gzip — still vastly under the 6.3MB parquet-wasm blob it replaced.fe2e832 -
Replace parquet-wasm + apache-arrow with hyparquet
The dataset's index parquets are tiny (~1–2 KB each) yet we were forcing every cold visitor to download a 6.3 MB WASM blob plus an Apache Arrow runtime to parse them. Swapped to hyparquet (pure-JS, ~30 KB, no init step) and stripped both heavy deps. parquet.ts now wraps the fetched ArrayBuffer as a synchronous AsyncBuffer; date columns surface as Date from hyparquet so we just normalize them to ISO strings here. First-paint payload before/after: - parquet_wasm_bg.wasm: 6,300 KB -> 0 (file no longer emitted) - apache-arrow chunk: ~140 KB -> 0 - hyparquet replacement: ~30 KB6e64fba -
Dedupe sharded manifest/rounds rows on overlap
Re-rendered batches occasionally produce a second shard whose rows collide on (match_id, map_name[, round]) — Svelte's each_key_duplicate fired on the homepage grid. Keep the freshest uploaded_at per key.a79c533
-
Initial CS2 render dataset viewer
31d3580