Danny Friar comments

Results 11 comments of


                                            Danny Friar

Avoid evaluating kernel when adding jitter

Related issues: - https://github.com/cornellius-gp/gpytorch/issues/1826 - https://github.com/cornellius-gp/gpytorch/issues/1951

Avoid evaluating kernel when adding jitter

The test failure is when calling the `predict` method of the deep GP multitask example. Possibly related to this issue https://github.com/cornellius-gp/gpytorch/issues/1892?

Avoid evaluating kernel when adding jitter

An example here that illustrates the issue clearly. ``` import torch import gpytorch from torch.profiler import profile, ProfilerActivity from torch.utils.data import DataLoader, TensorDataset N = 4567 M = 123 X...

Avoid evaluating kernel when adding jitter

@gpleiss were you able to take a look at this?

Avoid evaluating kernel when adding jitter

Thanks @gpleiss!

`clip` behaviour with NaN values is different between GPU and CPU onnx inference

@justinchuby sure I've added that now

[CLI]: `mlflow` importer uses old API that no longer exists

These docs also still reference the API described here https://github.com/wandb/wandb/issues/6184 which seems like it was removed a few months ago.

Block Matrix Linear Operator

> @gpleiss Thanks for the detailed review l. It all sounds good except for possibly the constructor which I need to think about. I am traveling right now but should...

Polars to_numpy slower with chunked array than going via pandas

That's interesting. Yeah I'm definitely running version `0.20.26`. I originally found this about 6 months ago and at the time I was unable to repro on a mac too. I...

Polars to_numpy slower with chunked array than going via pandas

![image](https://github.com/pola-rs/polars/assets/10010084/46287099-1e7c-4a98-a4a3-a6435699fe7e) This is what I get with `use_pyarrow=False`. Everything else is the same as above