Download legacy/tabular-sample/README.md from DiogenesLab/Diogenes_Multigame_Dev_Sessions_Archive_v01: direct link, hf CLI and curl.
- Browser
- Download file 3.3 kB
-
https://huggingface.co/datasets/DiogenesLab/Diogenes_Multigame_Dev_Sessions_Archive_v01/resolve/main/legacy/tabular-sample/README.md
- Command line
-
hf download hf://datasets/DiogenesLab/Diogenes_Multigame_Dev_Sessions_Archive_v01/legacy/tabular-sample/README.md
-
curl -L -o README.md https://huggingface.co/datasets/DiogenesLab/Diogenes_Multigame_Dev_Sessions_Archive_v01/resolve/main/legacy/tabular-sample/README.md
Historical multi-game tabular sample
This supplement preserves the original 23 structured files from the retired
DiogenesLab/Diogenes_Gameplay_multigame_sample_v01 distribution. All reachable
history was enumerated: eight commits contained one version of each of these
scientific files. These copies are byte-identical; original SHA-256 and byte
lengths are in SOURCE_MANIFEST.jsonl. Repository setup files and three outdated
dataset-card revisions are retained only in the private closeout audit.
The sample covers 603 segments from 19 development sessions / 19 games:
| File | Rows / purpose |
|---|---|
data/segments.parquet |
603 segment metadata, original media hashes, dimensions, clocks, eligibility and QC |
data/events.parquet |
332,572 keyboard/mouse input events, offsets and original capture timing |
data/frames.parquet |
205,091 frame/action records, timing, input state and capture defects |
mappings/*.json |
19 game-specific input-to-action mappings |
qc/qc_segments.json |
603 historical segment QC summaries |
Every segment is historically non_trainable=true and
production_certified=false. Failed QC, silent/degraded audio, short tails,
duplicate frames and input imbalance remain as recorded. These historical flags
are not changed by public availability. No split, training certification, or new
permission to underlying game IP is asserted; the archive's existing rights
notice applies. Input device values are opaque capture input handles, rather
than physical HID paths or names; session IDs and recording clocks are retained
for alignment.
Resolve media with the new index
The original video_path, audio_path, byte lengths and SHA-256 columns remain
historical claims about the original files. The old repository is retired.
Use MEDIA_INDEX.jsonl keyed by old_path to locate the current public file or
TAR member. Each row gives its pinned repository revision, public bytes and
SHA-256. Original and public hashes are both present and are never conflated.
There are 1,206 media objects: 1,140 retain original bytes, while the 34 Forza
Horizon 5 videos and 32 Halo Infinite videos are explicit privacy copies. Their
original unmasked files are held privately; public replacements preserve frame
times and decoded pixels outside the disclosed account/player-name masks.
See privacy/forza-horizon5/ and privacy/halo-infinite/ at the repository root
for transformation records and limits. Audio is unchanged. Historical frame/event
tables remain aligned to the preserved frame times; no generic replacement of
original media checksums has been performed.
import json
import pyarrow.parquet as pq
segments = pq.read_table("data/segments.parquet")
media = {r["old_path"]: r for r in map(json.loads, open("MEDIA_INDEX.jsonl"))}
first = segments.slice(0, 1).to_pylist()[0]
location = media[first["video_path"]]
print(location["public_location"], location["public_sha256"])
For public_location.kind == "direct_file", fetch the pinned path from the
specified Hugging Face dataset. For tar_member, fetch the pinned archive_path
and extract exactly member_path. Check the public SHA-256 and bytes in this
index, rather than the old table's original-byte hash when a privacy copy is used.