Closed geospith closed 9 years ago
How many iterations were run? Did it run to convergence? How does the objective grow/shrink?
I have the same problem in MF and I think it might be that LBFGS needs a different way of creating examples!?
Oddly LBFGS works in OptimizeByLearningSpecs.
Currently we don't use factorie optimization anymore. We could think about our own LBFGS implementation, but for now we can live with adagrad I think.
Batch optimisation with LBFGS runs, but returns nonsense results.