Enable more complex tests, fix related errors #392
Conversation
| A = jr.normal(akey, (10, 10), dtype=dtype) * 0.5 | ||
|
|
||
| if jnp.iscomplexobj(A) and isinstance(solver, diffrax.AbstractImplicitSolver): | ||
| return |
There was a problem hiding this comment.
What's the reason for this exclusion? I think ideally we do expect to be able to do this (once all the relevant upstream things have landed).
There was a problem hiding this comment.
Currently,
https://github.com/patrick-kidger/diffrax/blob/main/diffrax/_integrate.py#L638-L642
I agree that we will want to remove it in the future
There was a problem hiding this comment.
It used to work (somehow) when we merged the basic support, but now I think strict typing broke it in multiple places.
There was a problem hiding this comment.
I've added some related fixes (also see patrick-kidger/optimistix#53), but I think enabling it should come in a separate PR
patrick-kidger
left a comment
patrick-kidger
left a comment
There was a problem hiding this comment.
Okay, this LGTM! Let me know whether the points I've flagged up are intentional or not. If they are then I'm happy to merge this as-is.
|
|
||
| @pytest.mark.parametrize("solver_ctr,noise,theoretical_order", _solvers_and_orders()) | ||
| def test_sde_strong_order(solver_ctr, noise, theoretical_order): | ||
| @pytest.mark.parametrize("dtype", [jnp.float64]) |
There was a problem hiding this comment.
Is this intentionally leaving off complex support?
| @pytest.mark.parametrize( | ||
| "dtype", | ||
| (jnp.float64,), | ||
| ) |
|
|
||
| # Essentially used as a check that our general IMEX implementation is correct. | ||
| def test_sil3(): | ||
| @pytest.mark.parametrize("dtype", (jnp.float64,)) |
|
Yep all three are intentional since there are still failing tests and i prefer to fix them separately, keeping this pr only to type errors and test modifications. I'll open a separate PR for these |
|
Okay then! Great stuff, I've just merged this. |
* Fix complex tests * Fix more complex tests * New sde related fixes * New sde related fixes
2 participants
Similarly to patrick-kidger/lineax#89 many of fixes are just type casts due to strict typing. Should consider which of these need to be casts and which aren't
Also I included changes from #391 for simplicity