search

cornellius-gp / gpytorch

A highly efficient implementation of Gaussian Processes in PyTorch
MIT License
3.54k stars 557 forks source link

Issue with log likelihood calculation for straightforward multioutput GP example #2434

Open shubham0704 opened 11 months ago

shubham0704 commented 11 months ago

In the final cell of the code example that I have tried to create here: https://colab.research.google.com/drive/1Y_HhWtS9jGd-0GaLbDdjJixAtuvs5pDC?usp=sharing

image

There is a problem with likelihood calculation. I am unable to understand how to fix this given training is happening properly without hiccups.

Code Reference: https://docs.gpytorch.ai/en/latest/examples/03_Multitask_Exact_GPs/Batch_Independent_Multioutput_GP.html

kapusuzoglu commented 10 months ago

I think it is related to the fact that your test set is not providing the correct batch shape defined in BatchIndependentMultitaskGPModel.