# Inspect Viz

> Welcome to Inspect Viz, a data visualisation library for [Inspect
AI](https://inspect.aisi.org.uk/). Inspect Viz provides flexible tools
for creating high quality interactive visualisations from Inspect evaluations.

The the two most useful documentation sections are:

- The Docs section, which provides a comprehensive overview of how to create custom plots. Use it to determine which functions should be used for a given application. If the user is trying to create a new custom plot then this is a good place to start.
- The Examples section, which provides concrete examples of how to use the library. Reading this code is a good way to understand the most common features and usage patterns.

If you want to drill into specific functions, the Reference section provides comprehensive details on each function in the library.

Finally, these sections provide additional related documentation:

- The Views section provides documentation on pre-built plots for specific scenarios. 
- The Publishing section provides documetnation on how to publish visualizations as static images, notebooks, websites, and dashboards.

## Docs

- [Getting Started](https://meridianlabs-ai.github.io/inspect_viz/index.html.md): A high level overview of the entire library.
- [Plots](https://meridianlabs-ai.github.io/inspect_viz/components-plots.html.md): 
- [Marks](https://meridianlabs-ai.github.io/inspect_viz/components-marks.html.md): In depth guide to available marks.
- [Links](https://meridianlabs-ai.github.io/inspect_viz/components-links.html.md): How to direct links from visualizations to published Inspect log transcripts.
- [Tables](https://meridianlabs-ai.github.io/inspect_viz/components-tables.html.md): How to create tables that display an interactive grid of data used in your visualization. 
- [Inputs](https://meridianlabs-ai.github.io/inspect_viz/components-inputs.html.md): How to use inputs (e.g. sliders, select boxes, etc.) to create interactive visualisations.
- [Interactivity](https://meridianlabs-ai.github.io/inspect_viz/components-interactivty.html.md): How to create plot selections that enable filtering and cross-filtering of plot data based on inputs and interactors.

## Examples

- [Scores by Task](https://meridianlabs-ai.github.io/inspect_viz/examples/inspect/scores-by-task/index.html.md): Bar plot summarizing eval scores by model and task.
- [Scores by Factor](https://meridianlabs-ai.github.io/inspect_viz/examples/inspect/scores-by-factor/index.html.md): Bar plot for comparing eval scores by model and a boolean factor (e.g. no hint vs. hint).
| [Scores by Limit](https://meridianlabs-ai.github.io/inspect_viz/examples/inspect/scores-by-limit/index.html.md) | Line plot for evaluating how a model’s success rate changes as token or time limits  increase.
- [Scores Timeline](https://meridianlabs-ai.github.io/inspect_viz/examples/inspect/scores-timeline/index.html.md): Dot plot of eval scores for various models, with release date on the x-axis.
- [Scores Heatmap](https://meridianlabs-ai.github.io/inspect_viz/examples/inspect/scores-heatmap/index.html.md): Heatmap with values for comparing scores across model and task.
- [Scores by Model](https://meridianlabs-ai.github.io/inspect_viz/examples/inspect/scores-by-model/index.html.md): Visualise model scores on a single evaluation, with one or more baselines.
- [Tool Calls](https://meridianlabs-ai.github.io/inspect_viz/examples/inspect/tool-calls/index.html.md): Visualise tool usage over a series of turns.
- [Penguins Explorer](https://meridianlabs-ai.github.io/inspect_viz/examples/general/penguins/index.html.md): Dot plots, tables, color and symbol channels, filtering inputs.
- [Bias Parameter](https://meridianlabs-ai.github.io/inspect_viz/examples/general/bias-parameter/index.html.md): Area plots; slider inputs; channel expressions.
- [Seattle Weather](https://meridianlabs-ai.github.io/inspect_viz/examples/general/seattle-weather/index.html.md): Dot plots; bar plots; color and radius channels; x-interval selections, interactive legend. |
- [Athletes (Regression)](https://meridianlabs-ai.github.io/inspect_viz/examples/general/athletes-regression/index.html.md): Dot plots; regression lines; tables, filtering inputs, xy selections.
- [Athletes (Error Bars)](https://meridianlabs-ai.github.io/inspect_viz/examples/general/athletes-errorbars/index.html.md): Error bars; text annotations; params, slider inputs.

## Reference

- [inspect_viz](https://meridianlabs-ai.github.io/inspect_viz/reference/inspect_viz.html.md): Core API (data, components, params, selections).
- [inspect_viz.plot](https://meridianlabs-ai.github.io/inspect_viz/reference/inspect_viz.plot.html.md): Plot and legend creation.
- [inspect_viz.view](https://meridianlabs-ai.github.io/inspect_viz/reference/inspect_viz.view.html.md): Pre-built view functions for Inspect evaluation logs.
- [inspect_viz.mark](https://meridianlabs-ai.github.io/inspect_viz/reference/inspect_viz.marks.html.md): Plot marks (charts, annotations, etc.).
- [inspect_viz.interactor](ihttps://meridianlabs-ai.github.io/inspect_viz/reference/inspect_viz.interactor.html.md): Plot interactors for selection and navigation.
- [inspect_viz.transform](https://meridianlabs-ai.github.io/inspect_viz/reference/inspect_viz.transform.html.md):  Data transformers (aggregate, window, etc.).
- [inspect_viz.table](https://meridianlabs-ai.github.io/inspect_viz/reference/inspect_viz.table.html.md): Tabular data display.
- [inspect_viz.input](https://meridianlabs-ai.github.io/inspect_viz/reference/inspect_viz.input.html.md): Inputs for binding to parameters and selections.
- [inspect_viz.layout](https://meridianlabs-ai.github.io/inspect_viz/reference/inspect_viz.layout.html.md): Component layout functions.

## Views

- [scores_by_task()](https://meridianlabs-ai.github.io/inspect_viz/view-scores-by-task.html.md): Bar plot for comparing eval scores (with confidence intervals) across models and tasks.
- [scores_by_factor()](https://meridianlabs-ai.github.io/inspect_viz/view-scores-by-factor.html.md): Bar bar plot for comparing eval scores by model and a boolean factor (e.g. no hint vs. hint).
- [scores_by_limit()](https://meridianlabs-ai.github.io/inspect_viz/view-scores-by-limit.html.md): Line plot showing success rate by token limit.
- [scores_timeline()](https://meridianlabs-ai.github.io/inspect_viz/view-scores-timeline.html.md): Scatter plot with eval scores by model, organization, and release date. Filterable by evaluation and organization.
- [scores_heatmap()](https://meridianlabs-ai.github.io/inspect_viz/view-scores-heatmap.html.md): Heatmap with values for comparing scores across model and task.
- [scores_by_model()](https://meridianlabs-ai.github.io/inspect_viz/view-scores-by-model.html.md): Bar plot for comparing model scores on a single eval.
- [tool_calls()](https://meridianlabs-ai.github.io/inspect_viz/view-tool-calls.html.md): Heat map visualising tool calls over evaluation turns. 

## Publishing

- [Plots](https://meridianlabs-ai.github.io/inspect_viz/publishing-plots.html.md):  Publishing standalone plots as PNG images or HTML snippets.
- [Notebooks](https://meridianlabs-ai.github.io/inspect_viz/publishing-notebooks.html.md): Publish collections of plots and tables as a notebook.
- [Websites](https://meridianlabs-ai.github.io/inspect_viz/publishing-websites.html.md): Embed plots and tables in a larger website.
- [Dashboards](https://meridianlabs-ai.github.io/inspect_viz/publishing-dashboards.html.md): Create a dashboard to provide easy navigation of collections of plots.
- [PNG Output](https://meridianlabs-ai.github.io/inspect_viz/publishing-png-output.html.md): Rendering plots as PNG images within notebooks, websites, and dashboards.