Support arbitrary X in data
#100
Conversation
|
Just to link this to aleximmer/Laplace#144 |
Pull Request Test Coverage Report for Build 8791845160Warning: This coverage report may be inaccurate.This pull request's base commit is no longer the HEAD commit of its target branch. This means it includes changes from outside the original pull request, including, potentially, unrelated coverage changes.
Details
💛 - Coveralls |
|
Here's an example use case. https://curvlinops--100.org.readthedocs.build/en/100/basic_usage/example_huggingface.html The load for the users is minimal and the choice of |
There was a problem hiding this comment.
Did a first pass and added some minor suggestions, mainly missing documentation.
One main suggestion is to avoid having to duplicate the test functions, simply by appending the test cases using a dict to CASES_NO_DEVICE and instead make the cases fixture return the batch_size_fn.
|
One more thing, the type hints for the data also have to be modified in all files that don't require other changes, e.g. |
_ _
(_) ___ ___ _ __| |_
| |/ _/ / _ \/ '__ _/
| |\__ \/\_\/| | | |_
|_|\___/\___/\_/ \_/
isort your imports, so you don't have to.
VERSION 5.13.2
Nothing to do: no files or paths have have been passed in!
Try one of the following:
`isort .` - sort all Python files, starting from the current directory, recursively.
`isort . --interactive` - Do the same, but ask before making any changes.
`isort . --check --diff` - Check to see if imports are correctly sorted within this project.
`isort --help` - In-depth information about isort's available command-line options.
Visit https://pycqa.github.io/isort/ for complete information about how to use isort.
|
@f-dangel I'm done. Currently running the test on a GPU-enabled env. It takes so long but you can continue reviewing. PEP8 and other linters' issues should also be resolved. |
|
Alright, confirmed that all tests passed on GPU! |
|
@runame I'm done with my second pass, could you take a quick second look and merge if everything looks good? |
Closes #86
@f-dangel @runame please give feedback and answer the question below.
Assumption:
and there is an additional parameter in
_base._LinearOperator:where it must be non-
NonewheneverXis not atorch.Tensor.This also fits well with Huggingface, although one must replace HF's default dataloader to outputs
(data, data['labels'])instead of justdata. Let me know if this should be considered.Code for testing the functionality:
https://gist.github.com/wiseodd/426061afae24199446e60bfabc00e26e
I use
laplace-torchthere (two birds one stone), so just remove it if you don't want to install it. If you want to testlaplace-torch, install via