train_and_validate: Train and validate model on reported data

In incidental: Implements Empirical Bayes Incidence Curves

Description

This function fit models with selected hyperparameters on reported data and return a matrix of posterior Laplace samples.

Usage

train_and_validate(
  reported,
  delay_dist,
  lam,
  dof,
  beta0 = NULL,
  regularization_order = 2,
  reported_val = NULL,
  end_pad_size = 0,
  fisher_approx_cov = TRUE,
  num_samps_per_ar = 10
)

Arguments

reported	An integer vector of reported cases.
delay_dist	A positive vector that sums to one, which describes the delay distribution.
lam	A fixed value for the beta parameter regularization strength.
dof	Degrees of freedom for spline basis.
beta0	(optional) Initial setting of spline parameters (before optimization)
regularization_order	An integer (typically 0, 1, 2), indicating differencing order for L2 regularization of spline parameters. Default is 2 for second derivative penalty.
reported_val	Validation time series of equal size to reported vector for use with 'val' method. Default is NULL.
end_pad_size	And integer number of steps the spline is defined beyond the final observation.
fisher_approx_cov	A flag to use either the Fisher Information (TRUE) or the Hessian (FALSE) to approx posterior covariance over parameters.
num_samps_per_ar	An integer for the number of Laplace samples per AR fit.

Value

A list of results of train and validate, including:

• train_ll = training log likelihood

• val_ll = validation log likelihood (if 'reported_val' is not 'NULL')

• Isamps = samples of the incidence curve from a Laplace approximation

• Ihat = MAP estimate of the incidence curve

• Chat = expected cases given MAP incidence curve

• beta_hat = MAP estimate of spline parameters

• beta_cov = covariance of spline parameters

• beta_hess = Hessian of spline parameters

incidental documentation built on Sept. 16, 2020, 5:07 p.m.