Implementations of stochastic, limited-memory quasi-Newton optimizers, similar in spirit to the LBFGS (Limited-memory Broyden-Fletcher-Goldfarb-Shanno) algorithm, for smooth stochastic optimization. Implements the following methods: oLBFGS (online LBFGS) (Schraudolph, N.N., Yu, J. and Guenter, S., 2007 ), SQN (stochastic quasi-Newton) (Byrd, R.H., Hansen, S.L., Nocedal, J. and Singer, Y., 2016 ), adaQN (adaptive quasi-Newton) (Keskar, N.S., Berahas, A.S., 2016, ). Provides functions for easily creating R objects with partial_fit/predict methods from some given objective/gradient/predict functions. Includes an example stochastic logistic regression using these optimizers. Provides header files and registered C routines for using it directly from C/C++.
| github.com/david-cortes/stochQN | |
| Bug report | File report |