Bug fixes across Lee and swap to survival function for #243

[ljwolf]

This addresses a few bugs filed on the Lee statistic, as well as addressing the precision concerns raised in #243 with using 1-stats.norm.cdf().

[ljwolf]

Once merged, it would be good to make a bug fix release addressing this, given the lee standardisation issue can have an impact for odd weights specifications.

[codecov]

Codecov Report

Merging #245 (fe407b2) into main (165e139) will increase coverage by 1.5%.
The diff coverage is 72.7%.

@@           Coverage Diff           @@
##            main    #245     +/-   ##
=======================================
+ Coverage   71.5%   73.0%   +1.5%     
=======================================
  Files         24      24             
  Lines       3246    3246             
  Branches     519     519             
=======================================
+ Hits        2320    2369     +49     
+ Misses       763     709     -54     
- Partials     163     168      +5     

Impacted Files	Coverage Δ
esda/lee.py	72.8% <ø> (+53.3%)	⬆️
esda/geary.py	92.5% <33.3%> (ø)
esda/moran.py	74.5% <83.3%> (ø)
esda/getisord.py	66.4% <100.0%> (ø)

[ljwolf]

test failures arise in #244, so they are unrelated to these changes.

[ljwolf]

So the testing failures in the join counts is, I think, due to the changes upstream in libpysal having to do with adjacency lists.

[ljwolf]

OK, my understanding is the following.

With pysal/libpysal@main, the following is broken:

from libpysal.weights.util import lat2W

w = lat2W(3,3)
w.neighbors # this is correct
{0: [3, 1],
 3: [0, 6, 4],
 1: [0, 4, 2],
 4: [1, 3, 7, 5],
 2: [1, 5],
 5: [2, 4, 8],
 6: [3, 7],
 7: [4, 6, 8],
 8: [5, 7]}
w.to_adjlist().head() # this is not
   focal  neighbor  weight
0      0         3     1.0
1      0         4     1.0
2      3         0     1.0
3      3         1     1.0
4      3         2     1.0

The issue arises because, at line 440 of weights.py, we use self.neigbors.keys(). Since ids are sorted by default and dicts now retain their insertion order, the two are not the same:

w.id_order
[0, 1, 2, 3, 4, 5, 6, 7, 8]
w.neighbors.keys()
dict_keys([0, 3, 1, 4, 2, 5, 6, 7, 8])

The ordering that is needed is w.id_order, which matches the order in the sparse array we use for the edge tuples. With this libpysal fix, the test failures disappear.

[martinfleis]

Code-wise looks good. If someone close to the actual stats can have a look as well, it'd be good.

[sjsrey]

Looks good!