Fully runs a cox or fine-gray regression model with multiple column realizations, returning the model and results

CoxRunMulti uses a formula, data.table, and list of controls to prepare and run a Colossus cox or fine-gray regression function

Usage

CoxRunMulti(
  model,
  df,
  a_n = list(c(0)),
  keep_constant = c(0),
  realization_columns = matrix(c("temp00", "temp01", "temp10", "temp11"), nrow = 2),
  realization_index = c("temp0", "temp1"),
  control = list(),
  gradient_control = list(),
  single = FALSE,
  observed_info = FALSE,
  fma = FALSE,
  mcml = FALSE,
  cons_mat = as.matrix(c(0)),
  cons_vec = c(0),
  ...
)

Arguments

model

either a formula written for the get_form function, or the model result from the get_form function.

df

a data.table containing the columns of interest

a_n

list of initial parameter values, used to determine the number of parameters. May be either a list of vectors or a single vector.

keep_constant

binary values to denote which parameters to change

realization_columns

used for multi-realization regressions. Matrix of column names with rows for each column with realizations, columns for each realization

realization_index

used for multi-realization regressions. Vector of column names, one for each column with realizations. Each name should be used in the "names" variable in the equation definition

control

list of parameters controlling the convergence, see the Control_Options vignette for details

gradient_control

a list of control options for the gradient descent algorithm. If any value is given, a gradient descent algorithm is used instead of Newton-Raphson. See the Control_Options vignette for details

single

a boolean to denote that only the log-likelihood should be calculated and returned, no derivatives or iterations

observed_info

a boolean to denote that the observed information matrix should be used to calculate the standard error for parameters, not the expected information matrix

fma

a boolean to denote that the Frequentist Model Averaging method should be used

mcml

a boolean to denote that the Monte Carlo Maximum Likelihood method should be used

cons_mat

Matrix containing coefficients for a system of linear constraints, formatted as matrix

cons_vec

Vector containing constants for a system of linear constraints, formatted as vector

...

can include the named entries for the control list parameter

Value

returns a class fully describing the model and the regression results

See also

Other Cox Wrapper Functions: CoxCurveSolver(), CoxRun(), LikelihoodBound.coxres(), RunCoxRegression_Omnibus(), RunCoxRegression_Omnibus_Multidose()

Examples

library(data.table)
df <- data.table::data.table(
  "UserID" = c(112, 114, 213, 214, 115, 116, 117),
  "t0" = c(18, 20, 18, 19, 21, 20, 18),
  "t1" = c(30, 45, 57, 47, 36, 60, 55),
  "lung" = c(0, 0, 1, 0, 1, 0, 0),
  "dose" = c(0, 1, 1, 0, 1, 0, 1)
)
set.seed(3742)
df$rand <- floor(runif(nrow(df), min = 0, max = 5))
df$rand0 <- floor(runif(nrow(df), min = 0, max = 5))
df$rand1 <- floor(runif(nrow(df), min = 0, max = 5))
df$rand2 <- floor(runif(nrow(df), min = 0, max = 5))
names <- c("dose", "rand")
realization_columns <- matrix(c("rand0", "rand1", "rand2"), nrow = 1)
realization_index <- c("rand")
control <- list(
  "ncores" = 2, "lr" = 0.75, "maxiter" = 1,
  "halfmax" = 2, "epsilon" = 1e-6,
  "deriv_epsilon" = 1e-6, "step_max" = 1.0,
  "thres_step_max" = 100.0,
  "verbose" = 0, "ties" = "breslow", "double_step" = 1
)
formula <- Cox(t0, t1, lung) ~ loglinear(dose, rand, 0) + multiplicative()
res <- CoxRun(formula, df, control = control)