Renumics
Spotlight, developed by Renumics GmbH, is an open-source data curation tool designed specifically for unstructured data. The tool aims to e...
Last verified:
What is Renumics?
Renumics Spotlight is an open-source data curation tool designed for interactively exploring and understanding unstructured datasets. It helps machine learning and engineering teams quickly identify critical data segments, model failure modes, and curate high-quality datasets for training robust ML models. The tool works directly from Pandas dataframes, Hugging Face datasets, or Spotlight's HDF5 format, requiring just one line of code to visualize data.
Key features include support for multiple unstructured data types (images, audio, text, videos, time-series, and geometric data), interactive visualizations with embeddings/predictions/uncertainties, a unique Similarity Map for exploring embeddings, Audio-Player Widget with spectrograms, 3D Viewer for geometric data, object detection with overlay bounding boxes, customizable UI components (Inspector, Data Table, Scatter Plot, Histogram, Filter Bar), and Spotlight Pro for direct editing, curation, and annotation. The tool is built for data-centric AI workflows and enables cross-functional team collaboration.
Renumics Spotlight is designed for data scientists, machine learning engineers, and engineering teams working with complex unstructured data in industries like automotive, manufacturing, and engineering. It's particularly valuable for model validation, debugging, anomaly detection, root cause analysis, and data inspection workflows. The company also offers custom industrial AI services including workshops, data checks, proof of concepts, and end-to-end AI system development for test and simulation data analysis.
Renumics pricing
Pricing model: Free
Renumics Spotlight offers three pricing tiers. The Free Community Edition is open source (MIT License) and allows users to start analyzing unstructured datasets with basic visualization and exploration features. The Professional Edition costs 499€/month and enables cross-functional teams to curate high-quality datasets with Spotlight Pro features including direct editing, curation, and annotation for precise adjustments and granular curation. The Enterprise Edition is custom-priced (contact sales) and provides robust data-centric AI workflows tailored to specific use cases with full end-to-end support.
Renumics pros
- Open source with free community edition
- Works directly from existing Pandas dataframes
- One line of code to create visualizations
- Supports images, audio, text, videos, time-series, and geometric data
- Interactive Similarity Map for embedding exploration
- Audio-Player Widget with auto-generated spectrograms
- 3D Viewer for geometric data visualization
- Object detection with overlay bounding boxes
- Customizable UI components and layouts
- Data enrichments like embeddings, predictions, uncertainties
- Identifies critical data segments and model failure modes
- No vendor lock-in, modern tech stack
- Spotlight Pro enables direct editing and annotation
- Filter Bar for filtering data by feature values
- Integrates with Hugging Face datasets
- Lazy loading of unstructured data from disk or web storage
- MIT License, OSI approved
- Active open-source community contribution
Renumics cons
- Text support limited to single-line text widget
- Primarily designed for unstructured data, less optimal for tabular-only data
- Embedding concept mainly useful with images, text, video, audio
- Requires Python knowledge for installation and API usage
- Community edition lacks advanced curation features
- Pro edition costs 499€/month
- Web-based frontend requires browser access
- Custom layouts require initial configuration effort
Frequently asked questions about Renumics
How do I know that Renumics Spotlight is right for me?
Spotlight is right for you if you work with unstructured datasets (images, audio, video, text, time-series, or geometric data) and need to quickly understand, explore, and curate them. It's ideal for machine learning and engineering teams who want to identify critical data segments, find model failure modes, and build reliable ML models through better data curation. If you're currently spending significant time manually inspecting data or struggling to communicate about data issues with your team, Spotlight's interactive visualizations and collaborative features will help.
How do I install and get started with Spotlight?
Install Spotlight via pip with the command 'pip install renumics-spotlight'. Once installed, you can visualize your dataset with just one line of code: 'from renumics import spotlight' followed by 'spotlight.show(df)' where df is your Pandas DataFrame. You can also start Spotlight via the Command Line Interface (CLI) by specifying a file to load. Spotlight supports Pandas dataframes, Huggingface datasets, and Spotlight datasets based on the HDF5-Format.
What types of data does Spotlight support?
Spotlight supports most unstructured data types including images, audio, text, videos, time-series, and geometric data. For images, you can visualize them with object detection overlay bounding boxes. The Audio-Player Widget allows direct waveform interaction with auto-generated spectrograms. The 3D Viewer handles geometric data. While tabular data is supported, the tool is specifically optimized for unstructured or multimodal datasets that combine tabular metadata with unstructured content.
What is the difference between the free community edition and Spotlight Pro?
The free community edition provides core visualization and exploration capabilities for analyzing unstructured datasets. Spotlight Pro, available in the Professional Edition (499€/month), adds direct editing, curation, and annotation capabilities. Pro enables precise adjustments, granular curation, and better ML training through interactive data modification. The Enterprise Edition adds custom workflows and dedicated support for robust data-centric AI workflows.
How does the Similarity Map feature work?
The Similarity Map is a unique visualization component that lets you explore the interconnected landscape of your embeddings and features. It displays data points in a 2D space based on their embedding similarities, allowing you to identify clusters, outliers, and patterns in your dataset. You can integrate rich embeddings to enhance individual data points and transform model outputs into insightful features for deeper analysis through this interactive map.
Can I customize the Spotlight UI layout?
Yes, Spotlight provides a range of UI components that you can add, remove, and rearrange to create customized inspection layouts. Components include the Inspector and Data Table for analyzing individual datapoints, Similarity Map, Scatter Plot, and Histogram for understanding overall data distribution, and Filter Bar for filtering data based on feature values. Once you find a layout that works well, you can save it for future use and quickly load it without reconfiguring each time.
How does Spotlight handle large datasets?
Spotlight uses efficient in-memory analytics for tabular data (labels, metadata) which is loaded into memory when the dataset starts. Unstructured data samples (images, video, audio, time series) are loaded lazily from disk or web storage, meaning they're only loaded when needed for viewing. This approach allows Spotlight to handle large datasets efficiently without loading everything into memory at once.
Does Spotlight collect my data or send it to the cloud?
Spotlight注重用户隐私，仅收集性能统计数据，不收集个人敏感信息。The tool runs locally on your machine when you install it via pip, and your data stays on your system. Unstructured data is loaded lazily from your local disk or web storage that you specify, maintaining control over your data throughout the exploration process.
Can I use Spotlight with Hugging Face datasets?
Yes, Spotlight connects your stack to the open source ecosystem and specifically supports Hugging Face datasets. You can interactively explore Hugging Face datasets with just one line of code after installing Spotlight. The tool provides examples and layouts specifically designed for Hugging Face datasets, making it easy to analyze and visualize these datasets without conversion.
How does Spotlight help with model validation and debugging?
Spotlight helps identify critical data segments and model failure modes by leveraging data enrichments like embeddings, predictions, and uncertainties. You can use interactive visualizations to find where your model performs poorly, identify outliers and reference points, perform correlation and sensitivity analysis, and validate simulation studies. The Filter Bar lets you filter by prediction accuracy or other metrics to quickly find problematic data slices for model improvement.