Jeffrey Lai

I've just tried validating the implementation on blind cliffwalk and it seems much slower (~an order of magnitude) than the uniform replay buffer. The results below are for a one...

> To add to the minimum example shown above, it would be useful to have an interface for passing the auxiliary state to the vector field call of the dynamical...

@patrick-kidger It seems like standardization of this framework seems useful for a lot of problems. Do you think the `AuxSolver` template you've outlined above is enough as an initial version?...

@patrick-kidger With some testing, I'm unable to get your example of inheriting from `AbstractWrappedSolver` working, I'm unable to get around some module initialization bugs. However, the `def wrap_solver` method originally...

> It would be nice if the times at which auxiliary state(s) are saved could be specified by `SaveAt`, similar to how the ODE state is saved. Currently the [docs...

> Rather, I am suggesting that you should allocate a buffer at the start of the loop, put it in `solver_state`, and then `.at[].set()` into it. @patrick-kidger I thought to...