Auxiliary function for onlineVAR
fitting.
onlineVAR.control(lambda.ratio = NULL, nlambda = NULL,
lambda.min.ratio = NULL, abstol = 0.001, trace = FALSE, start = NULL,
parallel = FALSE, predall = FALSE)
Vector of penalization parameters as fractions of the
minimum lambda for which all coefficients are zero. If not specified
a sequence of lambda values is generated based on nlambda
and
lambda.min.ratio
. If supplied, nlambda
and
lambda.min.ratio
are ignored.
Number of lasso penalization parameters lambda. Default is 10.
Smallest value of lambda.ratio. Default is 0.0001
Absolute tolerance for coordinate descent convergence.
In each time step the algorithm stops when the sum of coefficient estimates
does not change more than abstol
. Default is 0.001.
If TRUE
coefficient estimates are stored for all time
steps. If FALSE
coefficient matrices are only stored for the last time
step to save memory.
Object of class "onlineVAR"
. Coefficient estimates from the
last time step are used as staring values. Can be used to continue updating
the model with new data.
If TRUE
the model fitting for the different lambda
is parallelized.
Logical whether predictions from all penalization parameters in the sequence are stored.
An list of components named as the arguments.
Number of lasso penalization parameters in the lambda sequence.
Smallest value for lambda.ratio.
Absolute tolerance for coordinate descent convergence.
Lambda sequence as fractions of the minimum lambda for which all coefficients are zero.
Logical whether coefficients should be stored for all time steps.
Starting values.
Logical whether the model fitting for the different lambda is parallelized.
Logical whether prediction from all penalization parameters in the sequence are stored.