Distance-based spatial weights in pysal

Question

I run into errors with distance-based spatial weights in pysal. When running a spatial lag model (e.g. twosls_sp.GM_Lag), I get the following exception:

Exception: All entries on diagonal must equal 0.

Trying the example here, the diagonal is 1 (i.e., an observation is also its own neighbor), and the output is different from that given in the docs:

import pysal
points=[(10, 10), (20, 10), (40, 10), (15, 20), (30, 20), (30, 30)]
w = pysal.DistanceBand(points,threshold=11.2,binary=False)
w.weights[0]
    [0.10000000000000001, inf, 0.089442719099991588]
w.neighbors[0]
    [1, 0, 3]

After row standardization, there are only zeros and nans in the weights matrix:

 w.transform = 'r'
 w.weights[0]
     [0.0, nan, 0.0]

Is this a bug in pysal, or likely a problem with my installation?

I am running the Enthought Canopy Python installation, with pysal installed separately, on a Mac OSX 10.8. Pysal is v1.11.0.

Answer 1

This is a bug that is currently being addressed.

Answer 2

The workaround I found is to use the insert_diagonal function. I'm still curious whether there is a problem with my pysal installation, however.

import pysal
import numpy as np
points=[(10, 10), (20, 10), (40, 10), (15, 20), (30, 20), (30, 30)]
w = pysal.DistanceBand(points,threshold=11.2,binary=False)
w.weights[0]
    [0.10000000000000001, inf, 0.089442719099991588]
w = pysal.weights.insert_diagonal(w, np.zeros(w.n))
w.weights[0]
    array([ 0.1       ,  0.08944272])
w.neighbors[0]
    [1, 3]

The new output is the same as the documented example, but I shouldn't need to manually modify the diagonal.