When the embedding has a number of zero eigenvalues, λ gets truncated.
However, U does not, and thus loadings throws a DimensionMismatch
error on master. This fixes that error.
Alternatively, we could truncate U as well, but retaining the full U
matrix makes it easier to obtain a full orthonormal basis, if desired.
When the embedding has a number of zero eigenvalues,
λgets truncated. However,Udoes not, and thusloadingsthrows aDimensionMismatcherror on master. This fixes that error.Alternatively, we could truncate
Uas well, but retaining the fullUmatrix makes it easier to obtain a full orthonormal basis, if desired.