plotting issue with splot.esda.lisa_cluster()

[jGaboardi]

Using splot.esda.lisa_cluster() to plot line-based LISA clusters seems to work with the most recent release of splot (v1.1.4). Here is a notebook demonstrating that the docstring example still works with polygons, but not with lines like it did previously.

[slumnitz]

@jGaboardi is there any error message that shows up?

[jGaboardi]

Nope 😕, no error message or warning. Just an empty plot (but with an intact legend).

[slumnitz]

ok I run into the same issue on macOS, executing lisa_cluster line by line I run into this error:

In [74]:     gdf.assign(cl=labels).plot(column='cl', categorical=True,
    ...:                                k=2, cmap=hmap, linewidth=0.1, ax=ax,
    ...:                                edgecolor='white', legend=legend,
    ...:                                legend_kwds=legend_kwds, **kwargs)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-74-7f905743daea> in <module>
----> 1 gdf.assign(cl=labels).plot(column='cl', categorical=True,
      2                            k=2, cmap=hmap, linewidth=0.1, ax=ax,
      3                            edgecolor='white', legend=legend,
      4                            legend_kwds=legend_kwds, **kwargs)

/usr/local/Caskroom/miniforge/base/envs/bio/lib/python3.8/site-packages/pandas/core/frame.py in assign(self, **kwargs)
   4479 
   4480         for k, v in kwargs.items():
-> 4481             data[k] = com.apply_if_callable(v, data)
   4482         return data
   4483 

/usr/local/Caskroom/miniforge/base/envs/bio/lib/python3.8/site-packages/geopandas/geodataframe.py in __setitem__(self, key, value)
   1320             except TypeError:
   1321                 warnings.warn("Geometry column does not contain geometry.")
-> 1322         super(GeoDataFrame, self).__setitem__(key, value)
   1323 
   1324     #

/usr/local/Caskroom/miniforge/base/envs/bio/lib/python3.8/site-packages/pandas/core/frame.py in __setitem__(self, key, value)
   3605         else:
   3606             # set column
-> 3607             self._set_item(key, value)
   3608 
   3609     def _setitem_slice(self, key: slice, value):

/usr/local/Caskroom/miniforge/base/envs/bio/lib/python3.8/site-packages/pandas/core/frame.py in _set_item(self, key, value)
   3777         ensure homogeneity.
   3778         """
-> 3779         value = self._sanitize_column(value)
   3780 
   3781         if (

/usr/local/Caskroom/miniforge/base/envs/bio/lib/python3.8/site-packages/pandas/core/frame.py in _sanitize_column(self, value)
   4502 
   4503         if is_list_like(value):
-> 4504             com.require_length_match(value, self.index)
   4505         return sanitize_array(value, self.index, copy=True, allow_2d=True)
   4506 

/usr/local/Caskroom/miniforge/base/envs/bio/lib/python3.8/site-packages/pandas/core/common.py in require_length_match(data, index)
    529     """
    530     if len(data) != len(index):
--> 531         raise ValueError(
    532             "Length of values "
    533             f"({len(data)}) "

ValueError: Length of values (303) does not match length of index (49)

whereby mask_local_auto produces 303 labels. Maybe this helps and we figure it out bit by bit? Does assign behave differently, did something change in geopandas/ pandas? @martinfleis

[jGaboardi]

Oh! You got an actual error message??

[slumnitz]

@jGaboardi I started debugging lisa_cluster line by line, so we need to resolve the missmatch between the assign and the gdf.

[jGaboardi]

Hmmm. I don't think anything changed within splot here for this release, right? So probably something a dependency?

[martinfleis]

Does assign behave differently, did something change in geopandas/ pandas?

I am not aware of any change on GeoPandas side regarding assign. Can you track it to a version?

[jGaboardi]

Versions are listed in the bug notebook:

spaghetti : 1.6.2
libpysal  : 4.5.1
esda      : 2.4.1
geopandas : 0.9.0
numpy     : 1.21.1
matplotlib: 3.4.2
splot     : 1.1.4
json      : 2.0.9
scipy     : 1.7.0

[martinfleis]

It is plotted, it is not empty. Just in white. This makes sense as we hard-code it as edgecolor='white'. If you remove that, it works.

splot/splot/_viz_esda_mpl.py

Lines 934 to 937 in 22039aa

	gdf.assign(cl=labels).plot(column='cl', categorical=True,
	k=2, cmap=hmap, linewidth=0.1, ax=ax,
	edgecolor='white', legend=legend,
	legend_kwds=legend_kwds, **kwargs)

I'd say that the code in splot is doing what it is supposed to now. Probably some plotting change in geopandas caused this but we need to fix it here.

[martinfleis]

Btw, I have no clue how did you get that error @slumnitz

[slumnitz]

we should also test how removing affects polygons instead of lines! Thank you @martinfleis!

[martinfleis]

I think we should treat LineStrings and Polygons differently under the hood and pass edgecolor to Polygons and not to LineStrings.

[slumnitz]

@jGaboardi, @darribas and @martinfleis what do you think about adding additional tests that warn the user if there is more than one geom_type in the gdf? Has this ever been an issue?

[martinfleis]

It is technically possible to have mixed type GeoDataFrame and still get contiguity and measure LISA, though I can't think of a use case that would make sense. We can certainly add a warning in that case but I am not worried about that very much.

[jGaboardi]

Resolving in #140.