STTLA PR #414
Open
STTLA PR #414
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
andyElking
force-pushed
the
sttla_pr
branch
2 times, most recently
from
7b9b1e7
to
a1f2e2a
Compare
**Major** - Renamed "bar"->"meter" to reflect the fact that e.g. the text output isn't a bar. - Added docstrings for all progress meter methods, the new `diffeqsolve(..., progress_meter=...)` argument, and added everything into the docs. - Fixed bug with t0!=0 producing the wrong results. - Fixed vmap'd diffeqsolves crashing progress bars. Making this work actually necessitated a fair bit of rewriting, with a gamut of different callbacks and unvmaps. **Minor** - Progress updates now move over `[0, 1]` rather than `[0, 100]`. - Tidied up tqdm bar formatting. - `TqdmProgressMeter` now checks at trace-time if `tqdm` is installed, rather than crashing at runtime. - The progress bar state is now a `PyTree[Array]`
* Added parametric control types * Demo of AbstractTerm change * Correct initial parametric control type implementation. * Parametric AbstractTerm initial implementation. * Update tests and fix hinting * Implement review comments. * Add parametric control check to integrator. * Update and test parametric control check * Introduce new LevyArea types * Updated Brownian path LevyArea types * Replace Union types in isinstance checks * Remove rogue comment * Revert _brownian_arch to single assignment * Revert _evaluate_leaf key splitting * Rename variables in test_term * Update isinstance and issubclass checks * Safer handling in _denormalise_bm_inc * Fix style in integrate control type check * Add draft vector_field typing * Add draft vector_field typing * Fix term test * Revert extemporaneous modifications in _tree * Rename TimeLevyArea to BrownianIncrement and simplify diff * Rename AbstractLevyReturn to AbstractBrownianReturn * Rename _LevyArea to _BrownianReturn * Enhance _term_compatiblity checks * Fix merge issues * Bump pre-commit and fix type hints * Clean up from self-review * Explicitly add typeguard to deps * Bump ruff config to new syntax * Parameterised terms: fixed term compatibility + spurious pyright errors Phew, this ended up being a complicated one! Let's start with the easy stuff: - Disabled spurious pyright errors due to incompatible between pyright and `eqx.AbstractVar`. - Now using ruff.lint and pinned exact typeguard version. Now on to the hard stuff: - Fixed term compatibibility missing some edge cases. Edge cases? What edge cases? Well, what we had before was basically predicated around doing ```python vf, contr = get_args(term_cls) ``` recalling that we may have e.g. `term_cls = AbstractTerm[SomeVectorField, SomeControl]`. So far so simple: get the arguments of a subscripted generic, no big deal. What this failed to account for is that we may also have subclasses of this generic, e.g. `term_cls = ODETerm[SomeVectorField]`, such that some of the type variables have already been filled in when defining it: ```python class ODETerm(AbstractTerm[_VF, RealScaleLike]): ... ``` so in this case, `get_args(term_cls)` simply returns a 1-tuple of `(SomeVectorField,)`. Oh no! Somehow we have to traverse both the filled-in type variables (to find that one of our type variables is `SomeVectorField` due to subscripting) *and* the type hierarchy (to figure out that the other type variable was filled in during the definition). Once again, for clarity: given a subscriptable base class `AbstractTerm[_VF, _Control]` and some arbitrary possible-subscripted subclass, we need to find the values of `_VF` and `_Control`, regardless of whehther they have been passed in via subscripting the final class (and are `get_args`-able) or have been filled in during subclassing (and require traversing pseudo-type-hierarchies of `__orig_bases__`). Any sane implementation would simply... not bother. There is no way that the hassle of figuring this out was going to be worth the small amount of type safety this brings... So anyway, after a few hours working on this *far* past the point I should be going to sleep, this problem this is now solved. This PR introduces a new `get_args_of` function, called as `get_args_of(superclass, subclass, error_msg_if_necessary)`. This acts analogous to `get_args`, but instead of looking up both parameters (the type variables we want filled in) and the arguments (the values those type variables have been filled in with) on the same class, it looks up the parameters on the superclass, and their filled-in-values on the subclass. Pure madness. (I'm also tagging @leycec here because this is exactly the kind of insane typing hackery that he seems to really enjoy.) Does anyone else remember the days when this was a package primarily concerned about solving differential equations? --------- Co-authored-by: Patrick Kidger <33688385+patrick-kidger@users.noreply.github.com>
- set reportUnnecessaryTypeIgnoreComment=true, as required by the static type "tests" in `test_term`; - remove now redundant, commented out, beatype based tests in `test_term`; - remove all unnecessary type ignore comments from across the codebase.
* Fix complex casting and types issues * Dependency version
* Fix complex tests * Fix more complex tests * New sde related fixes * New sde related fixes
* Enable implicit solvers for complex inputs * change version * make pyright happy
… pyright: ignore comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR adds space-time-time Levy area to
VirtualBrownianTreeandUnsafeBrownianPath. The changes to the overall logic are minor, I mostly just added the math for the space-time-time case.The additions to
test_brownian.pyare more extensive:test_conditional_statisticswe test the pvals of the marginals for W, H, and K separately. This confirms that they are all Gaussian, so their joint distribution is exactly determined by their means and covariances, which we also test for. All of these are conditioned on the sample just before and the one just after the current sample (as it was before). I moved the code for computing the statistics into theconditional_statisticsfunction.test_conditional_statisticsitself is based on thm 3.7 from the Single-seed paper, I chose to test its correctness from a different angle intest_whk_interpolation. This test tries three things: a) just final interpolation with no halving (brownian bridge) steps, b) only one halving step followed by a final interpolation, c) 15 halving steps (i.e.tol = 2**-15) and no final interpolation (aka. "zero" spline). Since they all have the same distribution as predicted by thm 3.7, I conclude that indeed everything is as it should be. This still relies on the assumption that the halving steps are correct (i.e. thm 3.5), but that was computed by James in his thesis, so I suppose that's a fair assumption :).In the third commit I added shape annotations to arrays in
AbstractBrownianIncrementand its subclasses. This is optional, so if you prefer we can remove this commit.In the fourth commit I just removed two unnecessary
# pyright: ignorecomments, so pyright doesn't complain. These have nothing to do with the rest of this PR.I hope this PR is less challenging to review. No hurry though :)