jaxtyping VS pytype

You want runtime typechecking.

See either beartype [1] or typeguard [2]. And if you're doing any kind of array-based programming (JAX or not), then jaxtyping [3].

[1] https://github.com/beartype/beartype/

[2] https://github.com/agronholm/typeguard

[3] https://github.com/patrick-kidger/jaxtyping

Try using [jaxtyping](https://github.com/google/jaxtyping).

It also supports numpy/pytorch/etc.

Since you mention ML use-cases, you might like jaxtyping.

jaxtyping: rich shape & dtype annotations for arrays and tensors (also supports PyTorch/TensorFlow/NumPy);

Heads-up that my newer jaxtyping project now exists.

As other folks have commented, type hints are now a big deal. For static typing the best checker is pyright. For runtime checking there is typeguard and beartype. These can be integrated with array libraries through jaxtyping. (Which also works for PyTorch/numpy/etc., despite the name.)

pytype (https://github.com/google/pytype) is based on symbolic interpretation of python bytecode, but with the data stack storing types rather than values. it works very well and has been running in production within google forever.

the nice thing about writing a bytecode rather than an AST interpreter is that you can leverage a lot of the work the python compiler does, and work with the resulting simplified code.

There is also the rarely mentioned pytype from Google, written in Python. And pyright from Microsoft is written in Typescript, pyre at Facebook in OCaml. Last time I checked, these had better type inference algorithms (Hindley-Milner?) than mypy.

https://github.com/google/pytype

my last job was working on [pytype](https://github.com/google/pytype), which uses abstract interpretation to do static type inference and checking for python. we used the cpython compiler to convert a program to bytecode, and then ran the bytecode through a VM where the abstract values were types. it worked extremely well, and could even typecheck completely unannotated code, though of course with less precision than when the user supplied some types.

extending that to runtime JIT compilation is an interesting idea; i'm not sure if any of the current JIT systems do that, but i don't see any reason it wouldn't be a useful technique, so they likely do.

some of them; others are internal tools used at google that the company is happy to let the devs open source, but which are not official google "products". my main project when I was at google was in that category: https://github.com/google/pytype - it was not an "official google product" in that google was not officially supporting it for external users, but it's an extensively used product within google and developing it was my full time job.

it's open source! check out https://github.com/google/pytype and https://github.com/google/pytype/blob/main/docs/developers/t... for more on the multi-file runner

Pytype checks and infers types for your Python code - without requiring type annotations. Pytype can catch type errors in your Python code before you even run it.

Pyre from Meta, pyright from Microsoft and PyType from Google provide additional assistance. They can 'infer' types based on code flow and existing types within the code.

we've written a little bit about what pytype does differently here: https://google.github.io/pytype/

our main focus is to be able to work with unannotated and partially-annotated code, and treat it on par with fully annotated code.

So, I tried out pytype the other day, and it was a not a good experience. It doesn't support PEP 420 (implicit namespace packages), which means you have to litter __init__.py files everywhere, or it will create filename collisions. See https://github.com/google/pytype/issues/198 for more information. I've since started testing out pyre.

What is the smart money doing for type checking in Python? I've used mypy which seems to work well but is incredibly slow (3-4s to update linting after I change code). I've tried pylance type checking in VS Code, which seems to work well + fast but is less clear and comprehensive than mypy. I've also seen projects like pytype [1] and pyre [2] used by Google/Meta, but people say those tools don't really make sense to use unless you're an engineer for those companies.

Am just curious if mypy is really the best option right now?

[1] https://github.com/google/pytype