i think it is actually a mix. yes SGD, but there's also online validation of hyperparameters via step-recorded search. Hope my citation is correct, i started forgetting things. [1]
AFAIK simulated annealing approach has been abandoned in favor of [1] [1] http://www.researchgate.net/publication/1916718_Recorded_Step_Directional_Mutation_for_Faster_Convergence On Wed, May 21, 2014 at 11:44 PM, Peng Zhang <[email protected]> wrote: > Namit, > > I think the theory behind Mahout’s logistic regression is stochastic > gradient descent, rather than maximum likelihood. > > Best Regards, > Peng Zhang > > > > On May 22, 2014, at 2:29 PM, namit maheshwari <[email protected]> > wrote: > > > Hello Everyone, > > > > Could anyone please let me know the algorithm used behind > > LogisticRegression in Mahout. Also AdaptiveLogisticRegression mentions an > > *annealing* schedule. > > > > I would be grateful if someone could guide me towards the theory behind > it. > > > > Thanks > > Namit > >
