A tool for trying and running various optimization strategies
Declare an objective function and optimize accross one or more algorithms
Currently implemented:
-
LBFGS – running a backtracking line search on the first iteration and stepping forward, following Wolfe conditions on subsequent iterations
-
SGD with optional momentum
-
Logistic regression – with optional regularization
-
Feature hashing – a la VW
-
Designed to run on functions of offline data or data represented in Spark RDDs