CarolinePascal 9f1807996b migration: compact non-contiguous episode indices before v3.0 convert ...

Datasets with deleted episodes (e.g. '*_clean' variants) keep gaps in
their episode numbering across data, videos, and metadata. The stock
v2.1->v3.0 converter renumbers data/videos by sorted file order (0..N-1)
but reads original gapped indices from episodes.jsonl, so it raises
'Number of episodes is not the same'. Add reindex_episodes(): when every
source agrees on the same (non-contiguous) episode set, remap it to
0..N-1 everywhere (data files + episode_index/index columns, per-camera
videos, episodes.jsonl, episodes_stats.jsonl, info.json) so conversion
succeeds. Verified end-to-end on danaaubakirova/svla_so100_task4_v3_clean
(gaps {20,37,38,39} -> 0..49).

2026-07-18 18:34:58 +02:00

..

classify.py

migration: honor matching SO joint names, convert leading SO block

2026-07-17 16:54:10 +02:00

extract_dataset.py

migration: add extract_dataset.py to pull a sub-dataset into a standalone repo

2026-07-17 15:45:29 +02:00

fix_dataset.py

migration: compact non-contiguous episode indices before v3.0 convert

2026-07-18 18:34:58 +02:00

prune_destination.py

migration: make prune deletions resilient to transient Hub timeouts

2026-07-17 21:21:00 +02:00

run_migration.py

migration: compact non-contiguous episode indices before v3.0 convert

2026-07-18 18:34:58 +02:00

slurm_migrate.py

migration: scope subfolder download, keep normalized joints as-is

2026-07-16 17:30:10 +02:00

so_arm_frame.py

migration: scope subfolder download, keep normalized joints as-is

2026-07-16 17:30:10 +02:00