Persist is a JupyterLab extension to enable persistent interactive outputs in JupyterLab notebooks. Basically, you can manipulate data either in Vega-Altair plots or in a data table and use the manipulated data later in code.

Check out the introductory video below.

This repository contains source code for Persist (PyPi) extension.

Visit the persist website for more examples and full documentation.

Watch on Youtube with CC

- JupyterLab >= 4.0.0 or Jupyter Notebook >= 7.0.0
- pandas >= 0.25
- altair >= 5
- ipywidgets
- anywidget

To install the extension, execute:

pip install persist_ext

If the Jupyter server was already running, you might have to reload the browser page and restart the kernel.

To remove the extension, execute:

pip uninstall persist_ext

Persist supports two types of interactive outputs — a custom data table and Vega-Altair (>=5.0.0, see requirements and caveats) charts. The following examples will walk you through creating each one. The examples are also available as notebooks in the examples folder of the repository. Each section will link to the corresponding notebook as well as a binder link for the notebook.

Persist currently works with pandas dataframes, so load/convert the data to pandas dataframe before using.

You can use the following code snippet to create a Persist-enabled interactive data table.

from vega_datasets import data # Load vega_datasets
import persist_ext as PR # Load Persist Extension

cars_df = data.cars() # Get the cars dataset as Pandas dataframe

PR.PersistTable(cars_df) # Display cars dataset with interactive table

Persist has a plotting module to create an interactive scatterplot or bar chart quickly. This module is a thin wrapper around Vega-Altair.

To create a scatterplot:

from vega_datasets import data # Load vega_datasets
import persist_ext as PR # Load Persist Extension

cars_df = data.cars() # Get the cars dataset as Pandas dataframe

PR.plot.scatterplot(data=cars_df, x="Miles_per_Gallon:Q", y="Weight_in_lbs:Q", color="Origin:N")

To create a barchart:

from vega_datasets import data # Load vega_datasets
import persist_ext as PR # Load Persist Extension

cars_df = data.cars() # Get the cars dataset as Pandas dataframe

PR.plot.barchart(data=cars_df, x="Cylinders:N", y="count()")

You can also use Vega-Altair charts directly by passing the chart object to the PersistChart function.

from vega_datasets import data # Load vega_datasets
import altair as alt
import persist_ext as PR # Load Persist Extension

cars_df = data.cars() # Get the cars dataset as Pandas dataframe

brush = alt.selection_interval(name="selection")

chart = alt.Chart().mark_point().encode(
    x="Weight_in_lbs:Q",
    y="Miles_per_Gallon:Q",
    color=alt.condition(brush, "Origin:N", alt.value("lightgray"))
).add_params(
    brush
)

PR.PersistChart(chart, data=cars_df)

Persist also supports composite Vega-Altair charts.

from vega_datasets import data # Load vega_datasets
import altair as alt
import persist_ext as PR # Load Persist Extension

movies_df = data.movies() # Get the cars dataset as Pandas dataframe

pts = alt.selection_point(name="selection", fields=["Major_Genre"])

rect = alt.Chart().mark_rect().encode(
    alt.X('IMDB_Rating:Q').bin(),
    alt.Y('Rotten_Tomatoes_Rating:Q').bin(),
    alt.Color('count()').scale(scheme='greenblue').title('Total Records')
)

circ = rect.mark_point().encode(
    alt.ColorValue('grey'),
    alt.Size('count()').title('Records in Selection')
).transform_filter(
    pts
)

bar = alt.Chart(width=550, height=200).mark_bar().encode(
    x='Major_Genre:N',
    y='count()',
    color=alt.condition(pts, alt.ColorValue("steelblue"), alt.ColorValue("grey"))
).add_params(pts)

chart = alt.vconcat(
    rect   circ,
    bar
).resolve_legend(
    color="independent",
    size="independent",
)

PR.PersistChart(chart, data=movies_df)

Persist works with Vega-Altair charts directly for the most part. Vega-Altair and Vega-Lite offer multiple ways to write a specification. However, Persist has certain requirements that need to be fulfilled.

• The selection parameters in the chart should be named. Vega-Altair's default behavior is to generate a name of the selection parameter with an auto-incremented numeric suffix. The value of the generated selection parameter keeps incrementing on subsequent re-executions of the cell. Persist relies on consistent names to replay the interactions, and passing the name parameter fixes allows Persist to work reliably.

• The point selections should have at least the field attribute specified. Vega-Altair supports selections without fields by using auto-generated indices to define them. The indices are generated in the source dataset in the default order of rows. Using the indices directly for selection can cause Persist to operate on incorrect rows if the source dataset order changes.

• Dealing with datetime in Pandas is challenging. To standardize the way datetime conversion takes place within VegaLite and Pandas when using Vega-Altair, the TimeUnit transforms, and encodings must be specified in UTC. e.g month(Date) should be utcmonth(Date).

Persist is developed as part of a publication and will appear in EuroVis 2024.

Supplementary material including example notebooks, walkthrough notebooks, notebooks used in the study (including participant notebooks) and the analysis notebooks can be accessed here.

Computational notebooks, such as Jupyter, support rich data visualization. However, even when visualizations in notebooks are interactive, they still are a dead end: Interactive data manipulations, such as selections, applying labels, filters, categorizations, or fixes to column or cell values, could be efficiently apply in interactive visual components, but interactive components typically cannot manipulate Python data structures. Furthermore, actions performed in interactive plots are volatile, i.e., they are lost as soon as the cell is re-run, prohibiting reusability and reproducibility. To remedy this, we introduce Persist, a family of techniques to capture and apply interaction provenance to enable persistence of interactions. When interactions manipulate data, we make the transformed data available in dataframes that can be accessed in downstream code cells. We implement our approach as a JupyterLab extension that supports tracking interactions in Vega-Altair plots and in a data table view. Persist can re-execute the interaction provenance when a notebook or a cell is re-executed enabling reproducibility and re-use.

We evaluated Persist in a user study targeting data manipulations with 11 participants skilled in Python and Pandas, comparing it to traditional code-based approaches. Participants were consistently faster with Persist, were able to correctly complete more tasks, and expressed a strong preference for Persist.

Persist uses hatch to manage the development, build and publish workflows. You can install hatch using pipx, pip or Homebrew (on MacOS or Unix).

Install hatch globally in isolated environment. We recommend this way.

pipx install hatch

Install hatch in the current Python environment.

WARNING: This may change the system Python installation.

pip install hatch

pip install hatch

Jupyter extensions use a custom version of yarn package manager called jlpm. When any relevant command is run, hatch should automatically install and setup up jlpm. After installing hatch with your preferred method follow instructions below for workflow you want. We prefix all commands with hatch run to ensure they are run in proper environments.

Run the setup script from package.json:

hatch run jlpm setup

When setup is completed, open three terminal windows and run the follow per terminal.

Setup vite dev server to build the widgets

hatch run watch_widgets

Start dev server to watch and build the extension

hatch run watch_extension

Run JupyterLab server with minimize flag set to false, which gives better stack traces aqnd debugging experience.

hatch run run_lab

To build the extension as a standalone Python package, run:

hatch run build_extension

To publish the extension, first we create a proper version. We can run any of the following

hatch version patch # x.x.1
hatch version minor # x.1.x
hatch version major # 1.x.x

You can also append release candidate label:

hatch version rc

Finally you can directly specify the exact version:

hatch version "1.3.0"

Once the proper version is set, build the extension using the build workflow.

When the build is successful, you can publish the extension if you have proper authorization:

hatch publish

The widget architecture of Persist is created using anywidget projects.

The interactive visualizations used by Persist are based on the excellent, Vega-Lite and Vega-Altair projects. Specifially the implementation of JupyterChart class in Vega-Altair was of great help in understanding how Vega-Altair chart can be turned into a widget. We gratefully acknowledge funding from the National Science Foundation (IIS 1751238 and CNS 213756).