There is still a corner case in which simple structure or perfect clustering is insufficient to get the number of zeros the user requests into a particular factor. In that case the code will (correctly) squash more coefficients to zero but the degrees of freedom will not reflect this. A real fix will involve rechecking the degrees of freedom in the create_FAobject methods for restrictions.1storder, restrictions.general, and restrictions.2ndorder
This problem also occurs when perfect clustering is requested. It is mostly fixed in the trunk; see
http:// codebrowse. launchpad. net/~bgokgm/ fair/trunk/ revision/ 4
There is still a corner case in which simple structure or perfect clustering is insufficient to get the number of zeros the user requests into a particular factor. In that case the code will (correctly) squash more coefficients to zero but the degrees of freedom will not reflect this. A real fix will involve rechecking the degrees of freedom in the create_FAobject methods for restrictions. 1storder, restrictions. general, and restrictions. 2ndorder