Download
HiPlot is an interactive visualization toolkit for exploring high-dimensional experiments, especially those produced during hyperparameter search or ablation studies. Its core view is a parallel-coordinates plot that lets you brush, filter, and highlight runs to spot trade-offs, correlations, and Pareto fronts at a glance. You can load results from simple CSV/JSON logs or programmatically push “experiments” with typed fields, metrics, and tags. The UI supports dynamic filtering, color mapping, and tooltip details so you can iteratively narrow to the most promising configurations. Because it renders as self-contained HTML, you can embed the visualization in notebooks, export it, or serve it as a lightweight web app for teammates. HiPlot also offers summary statistics, correlation hints, and outlier highlighting to surface patterns that aren’t obvious from raw tables.
Features
- Parallel-coordinates plots for dozens of metrics and hyperparameters
- Brushing, filtering, and color mapping for interactive subset analysis
- Simple data ingestion from CSV/JSON or programmatic experiment objects
- Exportable, self-contained HTML for sharing results
- Lightweight server and notebook integration for quick iteration
- Built-in stats, correlations, and outlier surfacing to guide decisions
Project Samples
