matplotlib / mpl-altair

Matplotlib Backend for Altair Visualization Library

Matplotlib Backend for Altair Visualization Library

Altair is a declarative statistical visualization library for Python, based on the Vega and Vega-Lite visualization grammars. This project converts Altair charts to Matplotlib objects that can be modified using the Matplotlib API, rendered as publication-qualuty figures, and integrated with Matplotlib based tools, libraries, and dashboards.

Contributing

Want to add docs? Features? Found a bug? Awesome!

Our style guide is matplotlib's and we use pytest for testing. Chat with us on gitter

Acknowledgments

Initial work on this project was supported by the John Hunter Matplotlib Summer Fellowship.

mpl-altair's Issues

Update freetype on travis

travis is using 2.6.1, is that the one we should be using?

axis

• ticks: locators/formatters
• legend information @palnabarun
• colorbar @kdorr

Add binning support

mpl-altair currently doesn't support conversion for Altair charts that use binning.

Code for reproduction

from vega_datasets import data
seattle = data.seattle_weather()
seattle.head()

Binning color in a scatter plot:

chart = alt.Chart(seattle).mark_point().encode(
    alt.X('wind'), alt.Y('temp_max'), alt.Color('temp_min', bin=alt.Bin(step=10))
)
mplaltair.convert(chart)

Binning in a bar chart:

chart = alt.Chart(seattle).mark_bar().encode(
    alt.X('wind', bin=True), alt.Y('count()')
)
mplaltair.convert(chart)

Actual outcome

In a chart that would work without binning, binning fails silently and the chart is converted as if binning is always false.

Binning color in a scatter plot:

Binning in a bar chart:

raises NotImplementedError until bar charts and aggregates are working.

Expected outcome
Converting a scatter plot with binned color should produce a chart that looks something like this:

Converting a binned bar chart should produce a chart that looks something like this:

Possible solution and notes

• Manipulate the data to be binned in parse_chart.ChannelMetadata().
• Binning is probably better to add when bar charts and aggregate functions are functional in mplaltair.
• Be aware that binning can happen in the encoding definition or in a transform.

For more info

• Altair's user guide on binning
• Vega-Lite documentation on binning

Include examples of code and output in readme

Distinguish between mark_point, mark_circle, and mark_square

mpl-altair does not distinguish between mark_point(), mark_circle(), and mark_square(). All three marks produce a scatter plot.

Implementation Notes

• Marks should be handled in _marks.py
• mark_point() should draw open circles by default
• mark_circle() should draw closed circles by default
• mark_square() should draw squares by default

For more information

• Altair user_guide

Categorical scatter plot

mpl-altair currently doesn't convert Altair charts that have categorical color, stroke, shape, or size encodings.

Code for reproduction

df = pd.DataFrame({
    'a': [1, 2, 3, 4], 'b': [7, 6, 5, 4], 'c': ['a', 'b', 'b', 'c']
})

chart = alt.Chart(df).mark_point().encode(
    alt.X('a'), alt.Y('b'), alt.Color('c:N')
)
mplaltair.convert(chart)

Actual outcome

A NotImplementedError is raised.

Expected outcome

Something that looks kind of like this:

Possible solutions

• Fix this on the Matplotlib side of things
• Incorporate a workaround similar to how mplaltair._marks._handle_line() deals with colors in line plots

Add timeUnit support

mpl-altair currently doesn't support conversion for Altair charts that use timeUnit aggregations.

Code for reproduction

This and other timeUnit transformations don't work:

from vega_datasets import data
seattle = data.seattle_weather()

chart = alt.Chart(seattle).mark_point().encode(
    alt.X('month(date)'), alt.Y('temp_max')
)
mplaltair.convert(chart)

Actual outcome

A NotImplementedError is raised for anything that has a timeUnit.

Expected outcome

Any plot with a timeUnit transformation should work.

This plot should look something like this:

Possible solution and notes

• Manipulate the data in parse_chart.ChannelMetadata() so that the data reflects the timeUnit aggregation. (see parse_chart.ChannelMetadata._handle_timeUnit().)
• Also note that timeUnit transformations can also be applied separately from timeUnit transformations in the encoding channel.

Build is failing

While working on the other PR, I saw that the master's Travis build is failing. On further investigations, I found that they were failing due to change in how Altair interpreted inline data. The earlier behavior was to keep Inline Data as it is, but now they convert it into Named Data. This was breaking our tests as we did not have support for that.

Ref:
Altair Changelog
Failing Travis Builds

Add aggregate support

mpl-altair currently doesn't support conversion for Altair charts that use aggregations.

Code for reproduction

data:

df = pd.DataFrame({
    'a': [1, 1, 2, 3],
    'b': [1.1, 2.2, 3.3, 4.4]
})

Example 1:

chart = alt.Chart(df).mark_point().encode(
    alt.X('a'), alt.Y('average(b)')
)
mplaltair.convert(chart)

Example 2:

chart = alt.Chart(df).mark_point().encode(
    alt.X('a'), alt.Y('count()')
)
mplaltair.convert(chart)

Actual outcome

A NotImplementedError is raised for anything that has aggregated data.

Expected outcome

Any plot with aggregation should work.

The first chart should look something like this:

The second chart should look something like this:

Possible solution

During the initial altair chart parsing, manipulate the data to include the aggregations. See parse_chart.ChannelMetadata() (specifically _aggregate_channel()).

Add bar charts

Bar charts aren't supported yet in mpl-altair.

Implementation Notes

• Marks should be handled in _marks.py
• Plotting different data types in a bar plot gives different types of bar charts in Altair (vertical, horizontal, skinny bars, fat bars).
• Plotting a categorical y against a categorical x gives a heatmap-like chart in Altair.
• Adding a color encoding does different things depending on the data type. Sometimes the chart gets stacked, sometimes it doesn't.

For more information

• Altair user_guide

create github page for mpl-altair

For now as a blog, eventually to host the documentation for the project.