Datasets:
|
Download README.md from THULab/record_test_3: direct link, hf CLI and curl.
- Browser
- Download file 2.59 kB
-
https://huggingface.co/datasets/THULab/record_test_3/resolve/main/README.md
- Command line
-
hf download hf://datasets/THULab/record_test_3/README.md
-
curl -L -o README.md https://huggingface.co/datasets/THULab/record_test_3/resolve/main/README.md
2.59 kB
| license: apache-2.0 | |
| task_categories: | |
| - robotics | |
| tags: | |
| - LeRobot | |
| - robotics | |
| - timeseries | |
| - tsfile | |
| - format:tsfile | |
| modality: | |
| - tabular | |
| - timeseries | |
| pretty_name: record-test_3 TsFile | |
| configs: | |
| - config_name: default | |
| data_files: | |
| - split: train | |
| path: data/record_test_3_train.tsfile | |
| size_categories: | |
| - 10K<n<100K | |
| # record-test_3 TsFile | |
| This dataset is a TsFile conversion of the LeRobot dataset | |
| [`sunparticle/record-test_3`](https://huggingface.co/datasets/sunparticle/record-test_3). | |
| The source dataset was created with [LeRobot](https://github.com/huggingface/lerobot) | |
| and is licensed under `apache-2.0`. | |
| ## Dataset Details | |
| - Source dataset: [`sunparticle/record-test_3`](https://huggingface.co/datasets/sunparticle/record-test_3) | |
| - Source format: LeRobot `v2.1` | |
| - Robot type: `so101_follower` | |
| - Modalities: Time-series robot state and action data | |
| - Source task: `place crown on capybara` | |
| - Split: `train` | |
| - Episodes: `40` | |
| - Frames / rows: `17403` | |
| - Tasks: `1` | |
| - Sampling rate: `30 fps` | |
| - Source videos: `40` | |
| - Converted data file: `data/record_test_3_train.tsfile` | |
| ## TsFile Layout | |
| The conversion writes one TsFile table for the train split: | |
| `record_test_3_train`. | |
| - `Time` is the TsFile time column in milliseconds. | |
| - `episode_index` and `task_index` are TAG columns. | |
| - `frame_index` and `sample_index` are scalar FIELD columns. | |
| - `action_0` through `action_5` are flattened action FIELD columns. | |
| - `observation_state_0` through `observation_state_5` are flattened observation FIELD columns. | |
| Vector column flattening preserves the source column name, replaces `.` with | |
| `_`, and appends the element index. For example, `observation.state` becomes | |
| `observation_state_0` through `observation_state_5`. | |
| ## Conversion Notes | |
| - `Time = round(timestamp * 1000)` and restarts per episode. | |
| - The source `timestamp` column is dropped because it is represented by `Time / 1000` seconds. | |
| - The source `index` column is renamed to `sample_index`. | |
| - No numeric time-series rows are dropped; the converted TsFile has `17403` rows. | |
| - Source `meta/` files are mirrored, with `meta/info.json` updated to describe the converted TsFile artifact. | |
| - Videos are not uploaded here. The source video feature `observation.images.front` | |
| and the `videos/` tree remain in the original dataset: | |
| <https://huggingface.co/datasets/sunparticle/record-test_3/tree/main/videos>. | |
| ## Read Example | |
| Use the Apache TsFile SDK or TsFile tools to read: | |
| ```text | |
| data/record_test_3_train.tsfile | |
| ``` | |
| For episode-level filtering, query the table with the TAG column, for example | |
| `WHERE episode_index = 0`. | |