Renumics/spotlight
Interactively explore unstructured datasets from your dataframe.
What it solves
Spotlight provides a way to quickly understand and explore unstructured datasets. It helps machine learning and engineering teams identify critical data clusters and find problematic data slices in complex datasets containing images, audio, text, videos, time-series, and geometric data.
How it works
It integrates with pandas DataFrames and Hugging Face datasets, allowing users to visualize unstructured data and its associated metadata (such as embeddings, predictions, and uncertainties) in a browser-based interface. Users can specify custom column types (e.g., spotlight.Image) or use pre-defined layouts to debug classification models and perform exploratory data analysis.
Who it’s for
Machine learning and engineering teams who need to analyze, debug, and communicate findings regarding unstructured data problems.
Highlights
- Supports multiple modalities: images, audio, text, video, time-series, and geometric data.
- Interactive browser-based visualization of dataframes.
- Integration with embeddings and model predictions for cluster identification.
- Pre-defined layouts specifically for model debugging and classification tasks.