Briefly, the issue is that IsotonicRegression takes (quite reasonably?) a 1D-array instead of a (n, 1) array as input.
(I ran into that issue while wrapping Tibshirani & Hoefling's near-isotonic regression (R package neariso) using rpy2.)