There should be tests that make sure that the results of the algorithms are sensible.
These don't necessarily need to be part of automated CI tests,
as mentioned in #40, but rather the examples.
I suggest to use a trivial case, for example, for DA:
5d, no dynamics, observe (directly) 1 component.
- posterior mean/cov should nearly match theoretical.
- test all of the methods
- test for different configs (ensemble size, iteration numbers, with and without localisation)
For optimisation, for example: a 5d quadratic objective
- Check optimum is found
- Check number of iterations not excessive.
There should be tests that make sure that the results of the algorithms are sensible.
These don't necessarily need to be part of automated CI tests,
as mentioned in #40, but rather the examples.
I suggest to use a trivial case, for example, for DA:
5d, no dynamics, observe (directly) 1 component.
For optimisation, for example: a 5d quadratic objective