Steve Schmerler

[MRG] DOC GPs: log_marginal_likelihood() log(theta) input

1

#### Reference Issues/PRs Fixes #24786 #### What does this implement/fix? Explain your changes. Document the fact that the input argument `theta` to GaussianProcess{Regressor,Classifier}'s `log_marginal_likelihood()` method is actually expected to be...

Limit `gita ll` column widths

2

Hi Thanks for this project. I have a question related to column widths in the `gita ll` output, using the current `master` (a5ba6fe). * long repo names shift the output...

Hi, I really like the idea behind the app, thanks for the project! I found that for roughly 50% of new bullet list items, typed text doesn't show up (see...

Serialization: `torch.load(..., map_location="cpu")` doesn't map all tensors in `Laplace` instance

1

Hi I came across an issue when serializing a `BaseLaplace`-derived object with tensors mapped to GPU and then reading it back in and mapping to CPU only. The same goes...

The line ```py crys.rpdf(coords, rmax=5.0, cell=cell, ...) ``` uses and older API that is no longer supported. It needs to be converted to using a `Structure` as input, where we...

laplace: default prediction settings

2

[This is a re-post of a question I had in a Helmholtz-internal UQ channel, where Nina Maria Gottschling made us aware of this package.] In [the docs](https://lightning-uq-box.readthedocs.io/en/latest/tutorials/regression/laplace.html) you mention that...

laplace: hyper-parameter optimization: is `sigma_noise` optimized?

2

[This is a re-post of a question I had in a Helmholtz-internal UQ channel, where Nina Maria Gottschling made us aware of this package.] In the [laplace package's regression example](https://github.com/aleximmer/Laplace/blob/main/examples/regression_example.py#L36-L44)...

Code duplications in `predictor.py`

1

https://github.com/mala-project/mala/blob/9b0c4fc45dc93c9e7165302d9cb0d56eb1e5fb32/mala/network/predictor.py#L37-L40 in `__init__` is the same as https://github.com/mala-project/mala/blob/9b0c4fc45dc93c9e7165302d9cb0d56eb1e5fb32/mala/network/predictor.py#L65-L68 in `predict_from_qeout()`. Since the former is in the constructor, I think the latter can be skipped. Also here https://github.com/mala-project/mala/blob/9b0c4fc45dc93c9e7165302d9cb0d56eb1e5fb32/mala/network/predictor.py#L105-L110 `self.data.grid_dimension = self.parameters.inference_data_grid`...

From `mala/datahandling/data_scaler.py`: `DataScaler.transform()` ``` def transform(self, unscaled): """ Transform data from unscaled to scaled. Unscaled means real world data, scaled means data as is used in the network. Data is...