Skip to content

Demomo-20191204/lux

 
 

Repository files navigation

A Python API for Intelligent Visual Discovery

Build Status Documentation Status Slack Binder CodeCov

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

Here are some slides from a recent talk on Lux.

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

Getting Started

To start using Lux, simply add an additional import statement alongside 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

Basic recommendations in Lux

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

Next-step Recommendations Based on User Context

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. Enhance Recommendations
  • Filter adds a filter to the current selection, while keeping attributes (on the X and Y axes) fixed. These visualizations shows 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 offered shows a roughly linear trend between the two variables. Filter Recommendations
  • 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 there are many colleges with average cost of around $20000 per year, corresponding to the bulge we see in the scatterplot view. Generalize Recommendations

See this page 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.

Easily exportable visualization object

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 from Tableau. The visualizations are rendered via Altair into Vega-Lite specifications.

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

Specified Visualization

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 intent languages for visualization 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
    differentRegions = VisList(["Region=?","AverageCost"],df)
    differentRegions

Example Vis List

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

Installation

To setup Lux manually for development purposes, you should clone the two Github repos for Lux: 1) the core Python Lux API and 2) the Jupyter widget frontend.

To install the Python Lux API:

    git clone https://github.com/lux-org/lux.git
    cd lux/
    python setup.py install

To install the Lux Jupyter Widget:

    pip install git+https://github.com/lux-org/lux-widget
    jupyter nbextension install --py luxWidget
    jupyter nbextension enable --py luxWidget

Dev Support

Lux is undergoing active development. Please report any bugs, issues, or requests through Github Issues or post on the #help channel in the Lux Slack org.

If you are interested in participating in a user study on Lux, please contact Doris Lee for more detail.

About

Python API for Intelligent Visual Data Discovery

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 99.9%
  • Other 0.1%