Spatial_Pearson issue

[rudyarthur]

from scipy import sparse
import libpysal
import esda

print(libpysal.__version__)
print(esda.__version__)

x = np.array([[1,1],
              [0,1]])
y = np.array([[0,1],
              [1,1]])
w = np.array([[0,1,1,0],
              [1,0,0,1],
              [1,0,0,1],
              [0,1,1,0]])
sw = sparse.csr_matrix(w)

from esda.lee import Spatial_Pearson

sp = Spatial_Pearson()
sp.fit(x,y)
print(sp.association_)

sp = Spatial_Pearson(sw)
sp.fit(x,y)
print(sp.association_)

The above with pysal version 4.7.0 and esda version 2.5.0 doesn't produce what it ought to. The first fit returns a 4x4 array where the docs say to expect a 2x2. The second crashes with "dimension mismatch" error.

Looking at the code in lee.py I think the x and y inputs need to be flattened. The default connectivity is the unit matrix which is probably never what is required in practice. I'd suggest requiring the user to input the spatial weight matrix as in the moran code.

Cheers, Rudy

[ljwolf]

Thanks for the report!

The input x and y arrays must be of shape (n,1). So, I think the flattening should happen in the user code.

We can definitely improve documentation and force the validation step to catch this, however!