Allow enabling/disabling state reinitialization for individual states

[dweindl]

So far, state-reinitialization can is only be switched on/off for all states (whose initial state depends on fixedParameters) at once.
Would be convenient, it this could be handled more dynamically by ignoring any reinitialization for the states for which x0 evaluates to NaN.

[dweindl]

Turns out my original plan doesn't work out as smoothly as hoped for (started on #1344). The inconvenience is, that one would have to add one additional constant to the model for each state (or group of states) that one may want to conditionally reinitialize or not.

Not completely sure how the ideal interface for that would look like. Would be good if one could enable/disable state reinitialization for individual states via ExpData, and without adding additional model parameters/constants. Any ideas comments on how to best handle that?

Ideally that would be also easily adaptable to more complex setups than preequilibration+simulation, where one might want to specify multiple (time-dependent) state reinitializations (independently of SBML events).

[paulstapor]

The problem is how the functions x0_fixedParameters and sx0_fixedParameters work, right?

[dweindl]

The problem is how the functions x0_fixedParameters and sx0_fixedParameters work, right?

Yes, the overall problem/motivation is that you can't use x0_fixedParameters to reinitialize varying sets of states. Either you reinitialize all states from x0_fixedParameters (with reinitializeFixedParameterInitialStates=true) or none (with reinitializeFixedParameterInitialStates=false), but nothing inbetween.

[paulstapor]

well, obviously one could hack this by saving the old state and wrapping around the current code in C++. But this is cumbersome and error-prone. Better would be to add an "ix" variable to the headers of those two functions/files, implementing a switch statement in the files themselves, like this is already done for sensitivity function files with ip or event function files now with ie. So, we already have a very comparable concept for similar cases...
Obviously, this would break Matlab vs Python version and we would need another "if (python_generated)" in AMICI... But that would be fine from my side... Implementable in 2 roughly days, I'd say, including tests and code review. Pure coding time should be less...

Btw.: This functionality would also be highly welcome when dealing with adjoint preequilibration...

[dweindl]

Oh, somehow I overlooked your reply @paulstapor . Sounds good, either some index list, or a boolean array. Maybe the former is better, as it's more in line with plist as you pointed out.

@FFroehlich - works for you? Any other suggestions?

[FFroehlich]

Sounds good!

[dweindl]

Just to make sure we are talking about the same thing. We would:

• add a list of state indices that are to be reinitialized after preequilibration to Model+ExpData (also for after presimulation?)
• change (s)x0_fixedParameters to initialize only those states whose index is on said list (or did you mean x0?)

For PEtab import that would mean that for constant initial values in the condition table, that for any state there, we would add a constant parameter that we would set to that value of the respective condition. If any condition has an estimated parameter there, we'd have to add a non-constant parameter there (or two, one for beforepreequilibration and one for afterwards).

[FFroehlich]

Just to make sure we are talking about the same thing. We would:

• add a list of state indices that are to be reinitialized after preequilibration to Model+ExpData (also for after presimulation?)
• change (s)x0_fixedParameters to initialize only those states whose index is on said list (or did you mean x0?)

For PEtab import that would mean that for constant initial values in the condition table, that for any state there, we would add a constant parameter that we would set to that value of the respective condition. If any condition has an estimated parameter there, we'd have to add a non-constant parameter there (or two, one for beforepreequilibration and one for afterwards).

Yes, sounds good!

[paulstapor]

Just to make sure we are talking about the same thing. We would:

• add a list of state indices that are to be reinitialized after preequilibration to Model+ExpData (also for after presimulation?)
• change (s)x0_fixedParameters to initialize only those states whose index is on said list (or did you mean x0?)

Should indeed be (s)x0_fixedParameters, I think..

For PEtab import that would mean that for constant initial values in the condition table, that for any state there, we would add a constant parameter that we would set to that value of the respective condition. If any condition has an estimated parameter there, we'd have to add a non-constant parameter there (or two, one for beforepreequilibration and one for afterwards).

Somehow not completely sure whether I got that...

[dweindl]

• add a list of state indices that are to be reinitialized after preequilibration to Model+ExpData (also for after presimulation?)

• change (s)x0_fixedParameters to initialize only those states whose index is on said list (or did you mean x0?)

How should that interact with the existing reinitializeFixedParameterInitialStates?

A) If reinitializeFixedParameterInitialStates == true, then all are reinitialized, otherwise only those on said list
B) If reinitializeFixedParameterInitialStates == true, then said index list is used, otherwise it's ignored.
C) Remove reinitializeFixedParameterInitialStates

A is the only backward compatible solution, but C seems more reasonable.