A simple way to explore your data through a Tableau-like interface directly in your Panel data applications.

panel-graphic-walker brings the power of Graphic Walker to your data science workflow, seamlessly integrating interactive data exploration into notebooks and Panel applications. Effortlessly create dynamic visualizations, analyze datasets, and build dashboards—all within a Pythonic, intuitive interface.

• Simplicity: Just plug in your data, and panel-graphic-walker takes care of the rest.
• Quick Data Exploration: Start exploring in seconds, with instant chart and table rendering via a Tableau-like interface.
• Integrates with Python Visualization Ecosystem: Easily integrates with Panel, HoloViz, and the broader Python Visualization ecosystem.
• Scales to your Data: Designed for diverse data backends and scalability, so you can explore even larger datasets seamlessly. (More Features Coming Soon)

This project is in its early stages, so if you find a version that suits your needs, it’s recommended to pin your version, as updates may introduce changes.

Install panel-graphic-walker via pip:

pip install panel-graphic-walker

Here’s an example of how to create a simple GraphicWalker pane:

import pandas as pd
import panel as pn

from panel_gwalker import GraphicWalker

pn.extension()

df = pd.read_csv("https://datasets.holoviz.org/windturbines/v1/windturbines.csv.gz", nrows=10000)

GraphicWalker(df).servable()

You can put the code in a file app.py and serve it with panel serve app.py.

In the GraphicWalker UI, you can save your chart specification as a JSON file. You can then open the GraphicWalker with the same spec:

GraphicWalker(df, spec="spec.json")

You may change the renderer to one of ‘explorer’ (default), ‘profiler’, ‘viewer’ or ‘chart’:

GraphicWalker(df, renderer='profiler')

In some environments, you may encounter message or client-side data limits. To handle larger datasets, you can offload the computation to the server or Jupyter kernel.

First, you will need to install extra dependencies:

pip install panel-graphic-walker[kernel]

Then you can use server-side computation with kernel_computation=True:

walker = GraphicWalker(df, kernel_computation=True)

This setup allows your application to manage larger datasets efficiently by leveraging server resources for data processing.

Please note that if running on Pyodide, computations will always take place on the client.

To learn more about all the parameters and methods of GraphicWalker, try the panel-graphic-walker Reference App.

• object (DataFrame): The data for exploration. Please note that if you update the object, the existing chart(s) will not be deleted, and you will have to create a new one manually to use the new dataset.
• field_specs (list): Optional specification of fields (columns).
• spec (str, dict, list): Optional chart specification as URL, JSON, dict, or list. Can be generated via the export method.
• kernel_computation (bool): Optional. If True, the computations will take place on the server or in the Jupyter kernel instead of the client to scale to larger datasets. The ‘chart’ renderer will only work with client side rendering. Default is False.

• renderer (str): How to display the data. One of ‘explorer’ (default), ‘profiler’, ‘viewer’, or ‘chart’. These correspond to GraphicWalker, TableWalker, GraphicRenderer, and PureRender in the graphic-walker React library.
• container_height (str): The height of a single chart in the viewer or chart renderer. For example, ‘500px’ (pixels) or ’30vh’ (viewport height).
• hide_profiling (bool): Whether to hide the profiling part of the ‘profiler’ renderer. Default is False. Does not apply to other renderers.
• index (int | list): Optional index or indices to display. Default is None (all). Only applicable for the viewer or chart renderer.
• page_size (int): The number of rows per page in the table. Only applicable for the profiler renderer.
• tab (‘data’ | ‘vis’): Set the active tab to ‘data’ or ‘vis’ (default). Only applicable for the explorer renderer. Not bi-directionally synced.

• appearance (str): Optional dark mode preference: ‘light’, ‘dark’, or ‘media’. If not provided, the appearance is derived from pn.config.theme.
• theme_key (str): Optional chart theme: ‘g2’ (default), ‘streamlit’, or ‘vega’. If using the FastListTemplate, try combining the theme_key ‘g2’ with the accent color

  #5B8FF9

  , or ‘streamlit’ and

  #ff4a4a

  , or ‘vega’ and

  #4c78a8

  .

• config (dict): Optional additional configuration for Graphic Walker. For example {"i18nLang": "ja-JP"}. See the Graphic Walker API for more details.

• clone: Clones the GraphicWalker. Takes additional keyword arguments. Example: walker.clone(renderer="profiler", index=1).
• chart: Clones the GraphicWalker and sets renderer="chart". Example: walker.chart(0).
• explorer: Clones the GraphicWalker and sets renderer="explorer". Example: walker.explorer(width=400).
• profiler: Clones the GraphicWalker and sets renderer="profiler". Example: walker.profiler(width=400).
• viewer: Clones the GraphicWalker and sets renderer="viewer". Example: walker.viewer(width=400).

• export_chart: Returns chart(s) from the frontend exported as either Graphic Walker Chart specification, vega-lite specification or SVG strings.
• save_chart: Saves chart(s) from the frontend exported as either Graphic Walker Chart specifications, vega-lite specification or SVG strings.
• export_controls: Returns a UI component to export the charts(s) and interactively set scope, mode, and timeout parameters. The value parameter will hold the exported spec.
• save_controls: Returns a UI component to export and save the chart(s) acting much like export_controls.

• add_chart: Adds a Chart to the explorer from a Graphic Walker Chart specification.
• calculated_field_specs: Returns a list of fields calculated from the object. This is a great starting point if you want to provide custom field_specs.

Our dream is that this package is super simple to use and supports your use cases:

• Great documentation, including examples.
• Supports your preferred data backend, including Pandas, Polars, and DuckDB.
• Supports persisting and reusing Graphic Walker specifications.
• Scales to even the largest datasets, only limited by your server, cluster, or database.

Other backends might be supported if they are supported by both Narwhals and PygWalker.

Via the backends example its possible to explore backends. In the data test fixture you can see which backends we currently test.

Contributions and co-maintainers are very welcome! Please submit issues or pull requests to the GitHub repository. Check out the DEVELOPER_GUIDE for more information.