[Bug]

[helvince]

🐛 Bug

Modifying the size of the root decomposition seems to have no effect.

To reproduce

** Code snippet to reproduce ** Modified colab notebook of SKIP example https://drive.google.com/file/d/14PzzJfpcbUAp1c8i1Tb5rNRY6-zbk6r_/view?usp=sharing

** Stack trace/error message **

// Paste the bad output here!

Expected Behavior

Setting covar_root_decomposition to true and varying gpytorch.settings.max_root_decomposition_size(X) should influence the model output. Currently it does not influence computation time and output is equivalent regardless of the root decomposition size.

System information

Please complete the following information:

• 1.9.1
• 1.13.1+cu116
• DISTRIB_ID=Ubuntu, DISTRIB_RELEASE=20.04 (Google Colab)

Additional context

I am new to the GPytorch framework so my apologies if I am not concise. I wish to do low-rank approximations of the inverse Kernel matrix by limiting the root decomposition size. However, I find the results and computation time to be unaffected by varying this setting. Is there something I am overlooking?

[gpleiss]

Please paste a code example here, rather than linking to a notebook file.