Model-predictive control

[rfaelens]

Basic implementation of model-predictive control. Requirements:

• mpc(tdmore, theta, suffix) theta is a named vector with the initial values. We should check whether these are covariates in tdmore. suffix is a suffix to add to the theta names. These should be outputs in the tdmore model. The tdmore model is adapted to have all parameters as IOV.
• estimate.tdmoreMpc performs an estimation using MPC. We do a for-loop going over all occasions, and performing an estimate each time. We adapt the covariates to include the updated THETA values. At the end, we adapt the covariates one last time so future predictions follow ipred. estimate.tdmoreMpc ignores the se.fit parameter, and does not calculate vcov

[rfaelens]

Some small things left to do:

• Check whether the required variables (suffix) are present as output
• Check whether the covariates supplied to estimate already contain the population theta's. If so, these cannot be adapted anymore.
• Make sure plot also correctly shows the population distibutions. (workaround: ipred$tdmore <- tdmoreBaseModel )
• Make estimate.tdmorefit work again
• Define what happens if you "re-estimate": in my opinion, this should only be allowed by providing NEW data. It is quite difficult however: what happens if you change the occasions in the regimen?