Consistent hot/cold spot colors for local moran

[weikang9009]

This PR is to

• fix inconsistent cold/hot spot colors in local moran scatterplot addressing Inconsistent colors for moran_scatterplot and lisa_cluster when p values are small #131 and BUG: plot_local_autocorrelation colors do not match between subplots #178
• plot the dots for the cold/hot spots in the local moran scatterplot after the insignificant ones so that the cold/hot spots are more visible

Before:

After:

[martinfleis]

Fine with this but we should probably move these plotting functions to esda anyway.

[jGaboardi]

Fine with this but we should probably move these plotting functions to esda anyway.

Agreed.

[knaaptime]

nice. I actually thought this was already fixed. The other thing we need to fix is ensuring a square axis. The current implementation always elongates the Y to make it look like the slope of the regression line is 1.

+1 on the move. We should probably start a plotting branch in esda soon

[martinfleis]

The other thing we need to fix is ensuring a square axis. The current implementation always elongates the Y to make it look like the slope of the regression line is 1.

You sure about that? right now the scale of units along x and y is the same. If you enforce square axis, this would get distorted.

[knaaptime]

yes. look at weis figure

the distance between 0 and 2 on y is not the same as on x. The slope of that line is not 1, but it looks like it is because the y is squished

[knaaptime]

checkout cell 6 in the splot docs https://nbviewer.org/github/pysal/splot/blob/main/notebooks/esda_morans_viz.ipynb

that slope is supposed to be .35

[martinfleis]

the distance between 0 and 2 on y is not the same as on x.

Sorry to disagree, but it is.

The slope is wrong but the axes and the scatter are correct.

They are not the same on the subplot within plot_moran but they are the same on moran_scatterplot.

So, I believe that:

1. on moran_scatterplot we need to fix the visualisation of the slope.
2. on plot_moran we need to ensure that scale of x and y is the same (ax.set_aspect('equal')) and fix the visualisation of the slope

[knaaptime]

lol ok yes, bad example

i'm still confident there's still an issue that all the moran plots in splot and up with a slope that appears to be 1

[martinfleis]

yeah, slope is wrong on all of these

[martinfleis]

I moved the discussion on regression line to #186. I have also figured out why it is wrong.