Function vikos::inert_stochastic_gradient_descent
[−]
[src]
pub fn inert_stochastic_gradient_descent<C, M, H>(cost: &C, start: M, history: H, learning_rate: M::Target, inertia: M::Target) -> M where C: Cost, M: Model<Target=C::Error>, H: Iterator<Item=(M::Input, M::Target)>
SGD tranining with constant learning rate and velocity
Velocity avoids being stuck on saddle points during optimization
A good default for inertia
is 0.9