Add option to round initial state

[stevencarlislewalker]

With some forms of simulation stochasticity, the numbers of individuals in each state needs to be rounded

• [ ] What is the specific use case?
• [ ] Priority is to figure out the priority of these use cases
• [ ] Is this even possible in TMB? We have had trouble, but we should try it out in the SIMULATE macro -- if it works in this context then that should be sufficient for our purposes because all of our use cases here relate to simulation/forecasting and not fitting/calibration

[stevencarlislewalker]

Having CppAD casting issues again -- reminiscent of issues we had with the eigenvector calculations.

[stevencarlislewalker]

Rounding is not working. Here is an explanation from @guanwg:

The reason behind the failure is that TMB considers the cast as cast from type 'CppAD::AD<CppAD::AD >' to type 'int' . CppAD::AD is a special type (different from typical types) in the sense that it not only stores a value of a variable but also keeps track of "symbol" of the variable through assignment and/or involvement in formula in order to do differentiation correctly.

We faced a similar problem with direct eigenvector computation, but worked around it by implementing a simple power method.

[bbolker]

Except for backward compatibility/testing purposes, I don't think rounding is terribly important. I would be willing to put this to one side for now. If/when we get to dealing with process error seriously we may need to treat state as continuous anyway.

[stevencarlislewalker]

I agree @bbolker. As we discussed, rounding is not crucial for a minimum viable product.

Although it does make testing much less straightforward.

@guanwg, let's stop any work on this issue for now.

[bbolker]

Maybe add a switch to base-MacPan to not round initial values, so you can do comparable runs?

[stevencarlislewalker]

This is a good idea @bbolker. I've explored this a bit, and am going to do it. Based on my initial poking around, there will still be a difference between the base-macpan and tmb-macpan. But this switch will make it easier to diagnose why.

[stevencarlislewalker]

options(MP_use_state_rounding = FALSE)

[stevencarlislewalker]

Perhaps we could do soft rounding here? But low priority in any case.