This function optimizes theta, in fact theta_weights. Since thetas are constrained (must be parameters of multinomial/discrete distribution), we don't directly optimize the likelihood function w.r.t. theta, but we perform change of variables to do unconstrained optimization. We therefore store these unconstrained variables in the field "theta_weights", and update these variables.
ISOpureS1.model_optimize.opt_theta(
tumordata,
model,
NUM_ITERATIONS_RMINIMIZE,
iter,
NUM_GRID_SEARCH_ITERATIONS
)a GxD matrix representing gene expression profiles of tumour samples
list containing all the parameters to be optimized
minimum number of iteration that the minimization algorithm runs
the iteration number
number of times to try restarting with different initial values
The model with the theta parameter updated