Python API for Intelligent Visual Data Discovery

Lux is a Python library that makes data science easier by automating aspects of the data exploration process. Lux facilitate faster experimentation with data, even when the user does not have a clear idea of what they are looking for. Visualizations are displayed via an interactive widget that allow users to quickly browse through large collections of visualizations directly within their Jupyter notebooks.

Here is a 1-min video introducing Lux, and slides from a more extended talk.

Try out Lux on your own in a live Jupyter Notebook here!

Getting Started

To start using Lux, simply add an extra import statement along with your Pandas import.

import lux
import pandas as pd

Then, Lux can be used as-is, without modifying any of your existing Pandas code. Here, we use Pandas’s read_csv command to load in a dataset of colleges and their properties.

    df = pd.read_csv("college.csv")
    df

Voila! Here’s a set of visualizations that you can now use to explore your dataset further!

Next-step recommendations based on user intent:

In addition to dataframe visualizations at every step in the exploration, you can specify in Lux the attributes and values you’re interested in. Based on this intent, Lux guides users towards potential next-steps in their exploration.

For example, we might be interested in the attributes AverageCost and SATAverage.

    df.intent = ["AverageCost","SATAverage"]
    df

The left-hand side of the widget shows the current visualization, i.e., the current visualization generated based on what the user is interested in. On the right, Lux generates three sets of recommendations, organized as separate tabs on the widget:

• Enhance adds an additional attribute to the current selection, essentially highlighting how additional variables affect the relationship of AverageCost and SATAverage. We see that if we breakdown the relationship by FundingModel, there is a clear separation between public colleges (shown in red) and private colleges (in blue), with public colleges being cheaper to attend and with SAT average of lower than 1400.
• Filter adds a filter to the current selection, while keeping attributes (on the X and Y axes) fixed. These visualizations show how the relationship of AverageCost and SATAverage changes for different subsets of data. For instance, we see that colleges that offer Bachelor’s degree as its highest degree show a roughly linear trend between the two variables.
• Generalize removes an attribute to display a more general trend, showing the distributions of AverageCost and SATAverage on its own. From the AverageCost histogram, we see that many colleges with average cost of around $20000 per year, corresponding to the bulge we see in the scatterplot view.

See this page for more information on additional ways for specifying the intent.

Easy programmatic access of exported visualization objects:

Now that we have found some interesting visualizations through Lux, we might be interested in digging into these visualizations a bit more. We can click on one or more visualizations to be exported, so we can programmatically access these visualizations further in Jupyter. Visualizations are represented as Vis objects in Lux. These Vis objects can be translated into Altair or VegaLite code, so that we can further edit these visualizations.

Quick, on-demand visualizations with the help of automatic encoding:

We’ve seen how Viss are automatically generated as part of the recommendations. Users can also create their own Vis via the same syntax as specifying the intent. Lux is built on the philosophy that users should always be able to visualize anything they want, without having to think about how the visualization should look like. Lux automatically determines the mark and channel mappings based on a set of best practices. The visualizations are rendered via Altair into Vega-Lite specifications.

    from lux.vis.Vis import Vis
    Vis(["Region=New England","MedianEarnings"],df)

Powerful language for working with collections of visualizations:

Lux provides a powerful abstraction for working with collections of visualizations based on a partially specified queries. Users can provide a list or a wildcard to iterate over combinations of filter or attribute values and quickly browse through large numbers of visualizations. The partial specification is inspired by existing work on visualization query languages, including ZQL and CompassQL.

For example, we might be interested in looking at how the AverageCost distribution differs across different Regions.

    from lux.vis.VisList import VisList
    VisList(["Region=?","AverageCost"],df)

To find out more about other features in Lux, see the complete documentation on ReadTheDocs.

Installation

To get started, Lux can be installed through PyPI.

    pip install lux-api

This installation includes the Lux Jupyter widget frontend, lux-widget. To activate the Jupyter notebook extension:

    jupyter nbextension install --py luxwidget
    jupyter nbextension enable --py luxwidget

If the installation happens correctly, you should see two - Validating: OK after executing the two lines above.

Note that Lux currently only works with Jupyter notebooks in Chrome. Support for Jupyter Lab will be coming soon. If you encounter issues with the installation, please refer to this page to troubleshoot the installation. Follow these instructions to set up Lux for development purposes.

Support and Resources

Lux is undergoing active development. If you are interested in using Lux, we would love to hear from you. Any feedback, suggestions, and contributions for improving Lux are welcome.

Other additional resources:

• Sign up for the early-user mailing list to stay tuned for upcoming releases, updates, or user studies.
• Visit ReadTheDoc for more detailed documentation.
• Clone lux-binder to try out these hands-on exercises or tutorial series on how to use Lux.
• Join our community Slack to discuss and ask questions.
• Report any bugs, issues, or requests through Github Issues.