New Data Type + Action: Automatic geographic attribute visualizations

[micahtyong]

In this PR

Closes #123 by adding a new data type, geographical, which detects columns representing geographic attributes (e.g. state, country), then calls the default action, univariate, which then aggregates quantitative attributes and displays the corresponding choropleth symbol map.

Changes

• Addition of new geographical datatype
• Modified the univariate default action and added SymbolMap visualization
• Utilizes existing aggregations from PandasExecutor.py
• Tests in tests/test_type.py, tests/test_action.py, and test_vegalite_default_actions_registered for correctness and failure cases

Example Output

[micahtyong]

cc: @dorisjlee. Just worked through this issue! Lux can now detect and visualize geographical attributes. I may have pulled upstream changes in a way s.t. this PR includes merged commits from other PRs, but the logic from those remains unchanged. Let me know if there are changes I might need to make. I hope this helps!

[micahtyong]

Hi @dorisjlee, I've added the ability to export choropleths, as requested. An example output is shown below. Let me know if there's anything else I should modify!

[dorisjlee]

Hi @micahtyong,
This looks really great! I tried this out on the Covid dataset that we worked on a while back and the maps look really cool with the different measure values in the dataset.

df = pd.read_csv("https://github.com/covidvis/covid19-vis/blob/master/data/interventionFootprintByState.csv?raw=True",index_col=0)
df['dateBefore'] = pd.to_datetime(df['dateBefore'], format='%Y-%M-%d')
df

One thing that I did notice from this dataset is that when we click on the visualization, there is a lot of "jitter" that happen with the plots being redrawn.

I believe this is related to lux-widget:#63. This re-drawing issue is present for all charts, but perhaps because the map vis takes longer to draw, so the jitter looks more off. This is a bigger issue that should be addressed separately, so I think we should be good to merge this issue independent of this.

I'll wait until @caitlynachen does a code review on this PR before merging this in.
Thanks!

[caitlynachen]

Looks great! Maybe next steps is to work on the same geo feature for Matplotlib, but amazing job figuring it out for Altair!

[dorisjlee]

Thanks @micahtyong! Feel free to open new issues on some of the next steps that we discussed for extending the geo vis features (e.g., matplotlib, etc).