admin/lerobot
1
0
Fork 0
mirror of https://github.com/huggingface/lerobot.git synced 2026-07-13 13:01:58 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(datasets): normalize shape=(1,) numeric values before HF encoding (#3344)

Browse Source 
* fix(datasets): normalize shape=(1,) numeric values before save

* test(datasets): cover shape=(1,) int/bool and finalize

Co-authored-by: Copilot <copilot@github.com>
This commit is contained in:
四七
2026-05-19 22:53:19 +08:00
committed by GitHub
parent d38eb89f71
commit 6a8878a639
2 changed files with 44 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/lerobot/datasets/dataset_writer.py
+8 -1
View File
@@ -250,7 +250,14 @@ class DatasetWriter:
for key, ft in self._meta.features.items():
if key in ["index", "episode_index", "task_index"] or ft["dtype"] in ["image", "video"]:
continue
episode_buffer[key] = np.stack(episode_buffer[key])
stacked_values = np.stack(episode_buffer[key])
# `shape=(1,)` numeric features are serialized as `datasets.Value`, which expects scalars.
# Normalizing to `(N,)` keeps save semantics stable across dependency versions.
if tuple(ft["shape"]) == (1,) and ft["dtype"] != "string":
stacked_values = stacked_values.reshape(episode_length)
episode_buffer[key] = stacked_values
# Wait for image writer to end, so that episode stats over images can be computed
self._wait_image_writer()