scoreopt.control: Tuning parameters for score optimisation by Stochastic...

In ProbReco: Score Optimal Probabilistic Forecast Reconciliation

Tuning parameters for score optimisation by Stochastic Gradient Descent

Description

Function to set tuning parameters for stochastic gradient descent used to find a reconciliation matrix that optimises total score. The defaults are those of \insertCiteadam;textualProbReco and more details on the tuning parameters can be found therein.

Usage

scoreopt.control(
  eta = 0.001,
  beta1 = 0.9,
  beta2 = 0.999,
  maxIter = 500,
  tol = 1e-04,
  epsilon = 1e-08
)

Arguments

eta	Learning rate. Deafult is 0.001
beta1	Forgetting rate for mean. Default is 0.9.
beta2	Forgetting rate for variance. Default is 0.999.
maxIter	Maximum number of iterations. Default is 500
tol	Tolerance for stopping criterion. Algorithm stops when the change in all parameter values is less than this amount. Default is 0.0001.
epsilon	Small constant added to denominator of step size. Default is 1e-8

References

\insertAllCited

See Also

Other ProbReco functions: inscoreopt(), scoreopt(), total_score()

Examples

#Change Maximum Iterations to 1000
scoreopt.control(maxIter=1000)

ProbReco documentation built on April 5, 2023, 5:10 p.m.