simode.control {simode}

Class containing control parameters for a call to simode

Description

Class containing control parameters for a call to simode

Usage

simode.control(
  optim_type = c("both", "im", "nls"),
  im_optim_method = c("BFGS", "Nelder-Mead", "CG", "L-BFGS-B", "SANN", "Brent"),
  nls_optim_method = c("BFGS", "Nelder-Mead", "CG", "L-BFGS-B", "SANN", "Brent"),
  im_optim_control = list(),
  nls_optim_control = list(),
  ode_control = list(method = "lsoda"),
  im_smoothing = c("splines", "kernel", "none"),
  im_grid_size = 0,
  bw_factor = 1.5,
  use_pars2vars_mapping = F,
  trace = 0,
  save_im_trace = F,
  save_nls_trace = F,
  obs_sets_fit = c("separate", "separate_x0", "together"),
  parallel = F,
  save_to_log = F,
  reg_alpha = -1,
  reg_pkg = c("glmnet", "ncvreg")
)

Arguments

optim_type

Controls what optimization will be performed: either only integral-matching ('im'), only nonlinear least squares ('nls') or both (the default, i.e., first integral-matching then nonlinear least squares starting from the integral-matching estimates).

im_optim_method

Method for optimization during the integral-matching stage. Accepted values are any method supported by the method argument in optim.

nls_optim_method

Method for optimization during the nonlinear least squares stage. Accepted values are the same as in im_optim_method.

im_optim_control

A list with control parameters for optimization during the integral-matching stage. Can include anything that would appear in the control argument in optim (depending on the choice of im_optim_method). See optim.

nls_optim_control

Control parameters for optimization during the nonlinear least squares stage (as in im_optim_control)

ode_control

A list with control parameters for the ODE solver. Can include the argument method appearing in the arguments to ode, as well as any other control parameters accepted as additional parameters in the call to ode.

im_smoothing

Choice of type of smoothing during the integral-matching stage (see Details).

im_grid_size

Number of points used in integral-matching grid (not relevant when im_smoothing='kernel'). Value <=0 means the grid size will be set according to maximum number of observations for any of the equations in the call to simode.

bw_factor

Controls the bandwidth when im_smoothing='kernel'. The bandwidth for each equation will be bw_factor*the maximum time interval between two observations (should be >= 1).

use_pars2vars_mapping

Whether to use pars2vars mapping (see Details).

trace

Report level (0-4), with higher values producing more tracing information (see Details).

save_im_trace

Whether to save trace information of integral-matching optimization, which can then be plotted using plot_trace.

save_nls_trace

Whether to save trace information of nonlinear least squares optimization, which can then be plotted using plot_trace.

obs_sets_fit

Controls the way multiple observation sets are fitted: either "separate" (each set can be fitted with its own parameter values and initial conditions), "separate_x0" (same parameter values fitted for all sets while initial conditions may be different for each set) or "together" (fitting the mean of all observations sets).

parallel

Controls whether to fit sequentially or in parallel multiple observation sets (obs_sets>1 in the call to simode) that are fitted separately (obs_sets_fit="separate"). Fitting in parallel requires that the parallel package will be installed. When running in parallel, output will not be displayed regardless of the trace level. Instead, one can set save_to_log to true to save the output to a log file.

save_to_log

Controls whether to redirect output to a log file. If true, output will be saved to the file 'simode.log' in tempdir.

reg_alpha

Value of tuning parameter alpha for regularization during the integral-matching stage. Negative value means no regularization. A value between 0 and 1 controls the type of regularization (see glmnet and ncvreg).

reg_pkg

What package to use for regularization (in case reg_alpha>=0).

Details

Possible values for im_smoothing are “splines” (the default), in which case smoothing will be performed using smooth.spline with generalized cross-validation, “kernel”, using own kernel smoother function, or “none” (using the observations as is, with interpolation if necessary). use_pars2vars_mapping controls whether to use a mapping of which equations are affected by each of the parameters. When set to true, previous matrices computed as part of the integral-matching estimation are stored during the integral-matching optimization, and are updated only for the equations that were affected by the change in the parameter estimates from the previous iteration. When the number of equations is large and some of the parameters affect only a few equations, setting this option to true can significantly reduce the optimization time during the integral-matching stage (while increasing the storage usage). This is especially true with derivative based optimization methods (such as “BFGS” of optim) which updates only one of the optimized parameters in each iteration. trace has 5 possible levels:
With trace=0, there would be no output displayed if there are no errors.
With trace=1, a message will be displayed at the beginning and end of each optimization stage.
With trace=2, non-critical errors occurring during the optimization iterations will be displayed.
With trace=3, non-critical warnings occurring during the optimization iterations will be displayed.
With trace=4, the calculated loss value for each iteration of the integral-matching and nonlinear least squares optimizations will be displayed.