The implementation of shrink relies on scipy, which I do not believe we want to make a dependency for building the docs.

See, for example, the double pendulum example which implements the code in python rather than using a scipy method which would provide better results.

Good observation! (Should probably not approve stuff with a heavy cold...)

It would be great to get rid of the scipy dependency.

The implementation of shrink relies on scipy, which I do not believe we want to make a dependency for building the docs.

Would it be okay if the example uses "shrink=None" so that importing scipy is bypassed?
I think the shrink option gives a more pleasing result, and it would be good if we can let the users play with it.
If this is not acceptable, I can simply remove the shrink option and its dependency on scipy.

Welcome back @leejjoon !

I think the effect you are looking for can be had with library-native changes to the explode parameter rather than passing through scipy.ndimage.

Here is a quick example demonstrating the differences between the example as is (with your fixes so that it works, but without shrink, with your shrink implementation and with no shrink, but a uniform 0.1 for explode rather than (0, 0.05, 0, 0)

Okay, I have removed shrink parameter so the scipy dependency. However, I still think that it would be good to have a shrink parameter as it offers a rather general solution not specific to pie chart.

I also have rebased the PR to the current main and force-pushed it.

Welcome back @leejjoon !

Thanks!