It is possible that the answer (the final solution vector x) given by two different algorithms (such as the one in mllib and in R) are different, as the problem may not be strictly convex and multiple global optimum may exist. However, these answers should admit the same objective values. Can you give an example such that the objective value of other method is better (smaller) than the obj of mllib?
2014-07-27 11:06 GMT-07:00 Aureliano Buendia <buendia...@gmail.com>: > Hi, > > The recently added NNLS implementation in MLlib returns wrong solutions. > This is not data specific, just try any data in R's nnls, and then the same > data in MLlib's NNLS. The results are very different. > > Also, the elected algorithm Polyak(1969) is not the best one around. The > most popular one is Lawson-Hanson (1974): > > http://en.wikipedia.org/wiki/Non-negative_least_squares#Algorithms > > >
