Renumics-spotlight

Fixes

- Fix usage of boolean, integer and float `pandas.DataFrame` columns with new `numpy` versions.

Features

- Automatically create embeddings for columns containing images (ViT), audio (Wav2Vec2), or strings (gte-base)
python
spotlight.show(..., embed=['colA', 'colB'])

- New DType for axis aligned bounding boxes (`bounding_box_dtype`).
- New DType for Sequences of values. Supports bounding boxes and categories for now.
- New lens for labeled axis-aligned bounding boxes on images.
- New metrics: Levensthein distance, BLEU Score.
- New lenses for text: ROUGE & BLEU Score.

Features

- Add lens for visualizing axis-aligned bounding boxes with labels on images
- Create automatic embeddings for chosen columns (models `ViT`, `Wav2Vec2` and `GTE`) (parameter `embed_columns` in show; parameters `--embed-all` and `--embed` in CLI)

Fixes

- Do not use `take` function of `pyarrow.Table` as lon as it is bugged (s. https://github.com/apache/arrow/issues/25822 and https://issues.apache.org/jira/browse/ARROW-9773 for more)

Fixes

- Fixed a problem when reading large binary data (e.g. audio files or images) from HuggingFace datasets.

Features

- Toggle scientific notation for numbers in global settings
- Set lens settings through Python Layout API
- Support embedding dtype length in Spotlight Dataset
- Support axis-aligned bounding boxes in Spotlight Dataset

Fixes

- Keyboard scroll in table widget
- Fix crash on external data in H5 datasets

Features

- Lens Settings are persisted in Backend.

Fixes

- Fixed infinite loading bug when fetching lazy lens values.