Online selection rules: the meta-learning rows: private API
PortfolioOptimisers.rate_grid_experts — Function
rate_grid_experts(eta_min::Real, K::Integer, obj::AbstractOnlineObjective)Builds the K GradientProjection experts of Ader and Sword on a geometric grid of rates, each on the objective obj.
Every rate takes the numeric type of eta_min.
Mathematical definition
\[\begin{align} \eta_k &= 2^{k - 1} \eta_{\min}\,, \quad k = 1, \ldots, K\,. \end{align}\]
Where:
- $\eta_k$: Grid rate of expert $k$, the learning rate of its gradient step.
- $\eta_{\min}$: Smallest grid rate, the rate of the first expert.
- $K$: Expert count of an expert mixture.
The largest rate is $2^{K - 1}$ times the smallest.
Arguments
eta_min: The smallest rate of the grid.K: The number of experts.obj: The objective that every expert steps on.
Validation
eta_min > 0,K >= 1. ADomainErroris thrown otherwise.
Returns
experts::Vector{<:MirrorDescent}: TheKexperts. Expert $k$ takes the Euclidean gradient step at the rate $\eta_k$.
Related