bma_ens_models: Fit BMA coefficients for beta distributions with discrete...

In kdayday/forecasting: Probabilistic Solar Forecsting

Description

Fit BMA coefficients for beta distributions with discrete component at 1 for each ensemble member Data should be pre-processed and normalized to [0,1] so that clipped values are exactly 1

Usage

bma_ens_models(tel, ens, bma_distribution = "beta", max_power = NA,
  lr_formula = y ~ x, A_transform = NA, lm_formula = y ~ x + 0,
  B_transform = NA, percent_clipping_threshold = 0.995, tol = 0.001,
  ...)

Arguments

tel	Vector of training telemetry data on [0,1] OR a matrix [time x member] of training telemetry data tailored to each member
ens	Matrix of training ensemble member data [time x member] on [0,1]
bma_distribution	Either "beta" or "truncnorm" to select type of distribution for member kernel dressing
lr_formula	Formula in terms of x,y for logistic regression model, defaults to "y ~ x". Requires a negative x intercept to model PoC < 0.5.
A_transform	A function for transforming forecast data before logistic regression to get a's (optional)
lm_formula	Formula in terms of x,y for linear regression model, defaults to "y ~ x + 0"
B_transform	A function for transforming forecast data before linear regression to get b's (optional)
percent_clipping_threshold	[0,1] Power is designated as clipped when above this percentage of the maximum power
tol	A tolerance for determining if normalized values all are <=1 (defaults to 0.001)
...	Optional arguments to pass to EM algorithm

Value

A list for a discrete-continuous mixture model with beta distribution

kdayday/forecasting documentation built on Oct. 7, 2020, 7:16 p.m.