ianhill60 comments

Results 9 comments of


                                            ianhill60

[Question] Implementing multi-output multi-task approximate GP

I have a very similar question: Does Gpytorch support batching for correlated multitask regression? Would you use a MultitaskVariationalStrategy to do that? I stumbled across this tds article that may...

I'd like to request batch-training support for the MultitaskGPModel. Here is my code attempting to integrate a DataLoader to train a MultitaskGPModel. Unfortunately, I receive the RuntimeError: You must train on the training inputs! [Feature Request]

Thanks for getting back! Does my error have to do with how I induce the model? Here is my code and a data sample for testing: https://github.com/ianhill60/gprsampledataset/blob/7de2aa26288e6bd95d1e5772fcab81dc1d165b6a/multitaskBatched.py https://github.com/ianhill60/gprsampledataset/blob/a52c97f45ddb065dac8e9e32b2ce54c7afec3a02/xdf.csv https://github.com/ianhill60/gprsampledataset/blob/a52c97f45ddb065dac8e9e32b2ce54c7afec3a02/ydf.csv The...

I'd like to request batch-training support for the MultitaskGPModel. Here is my code attempting to integrate a DataLoader to train a MultitaskGPModel. Unfortunately, I receive the RuntimeError: You must train on the training inputs! [Feature Request]

Thanks for the helpful response wjmaddox! I applied this approach to my problem, but am still running into memory overflow errors even with batching. https://github.com/ianhill60/gprsampledataset/blob/69359be7f70c4325f004d2cbcb44ebd4d1fca1dc/VariationalMultitaskApplication.py What am I missing? I'm...

I'd like to request batch-training support for the MultitaskGPModel. Here is my code attempting to integrate a DataLoader to train a MultitaskGPModel. Unfortunately, I receive the RuntimeError: You must train on the training inputs! [Feature Request]

Thanks again! I was able to get this to run with a dataloader and batching, however my loss is not decreasing. It fluctuates between around 3.9 - 4.1 regardless of...

I'd like to request batch-training support for the MultitaskGPModel. Here is my code attempting to integrate a DataLoader to train a MultitaskGPModel. Unfortunately, I receive the RuntimeError: You must train on the training inputs! [Feature Request]

Thanks gpleiss! I implemented kmeans to learn inducing locations and used 1000 points. Model results improved but asymptote quickly (the loss starts at about 7, gets to a loss of...

I'd like to request batch-training support for the MultitaskGPModel. Here is my code attempting to integrate a DataLoader to train a MultitaskGPModel. Unfortunately, I receive the RuntimeError: You must train on the training inputs! [Feature Request]

Thank you gpleiss! I noticed that increasing the number of latent functions learned to correlate tasks increased my prediction accuracy and need for more data. Is there a 'right' number...

ianhill60

[Question] Implementing multi-output multi-task approximate GP

I'd like to request batch-training support for the MultitaskGPModel. Here is my code attempting to integrate a DataLoader to train a MultitaskGPModel. Unfortunately, I receive the RuntimeError: You must train on the training inputs! [Feature Request]

I'd like to request batch-training support for the MultitaskGPModel. Here is my code attempting to integrate a DataLoader to train a MultitaskGPModel. Unfortunately, I receive the RuntimeError: You must train on the training inputs! [Feature Request]

I'd like to request batch-training support for the MultitaskGPModel. Here is my code attempting to integrate a DataLoader to train a MultitaskGPModel. Unfortunately, I receive the RuntimeError: You must train on the training inputs! [Feature Request]

I'd like to request batch-training support for the MultitaskGPModel. Here is my code attempting to integrate a DataLoader to train a MultitaskGPModel. Unfortunately, I receive the RuntimeError: You must train on the training inputs! [Feature Request]

I'd like to request batch-training support for the MultitaskGPModel. Here is my code attempting to integrate a DataLoader to train a MultitaskGPModel. Unfortunately, I receive the RuntimeError: You must train on the training inputs! [Feature Request]

Sparse, Binary Data: Interpolate Missing

Sparse, Binary Data: Interpolate Missing

UnicodedecodeError