Regression Learner Slow to STOP

11 views (last 30 days)
Nelson Chow
Nelson Chow on 26 Aug 2020
Commented: Nelson Chow on 2 Sep 2020
Hi Guys,
Imagine going to a Ramen place, you order a ChaShu Ramen, the kitchen started cooking it, you waited for 18 hours. No Ramen. So you asked the order to be cancelled and asked for the bill. the bill comes after hours of waiting... this is the problem
I am NOT saying that the training process is slow, training of model takes time, that depends on your data sizes, algorithm etc. No problem it takes forever to train. Take however long you need to cook that Space Shuttle Ramen, I get that, I have ordered some hard things....
My problem is how slow Matlab is to STOP training.
e.g. in a Gaussian Process Regression, when i decied to STOP the training by hittling the cross circle, the Learner take ONE HOUR to stop. It just has a spinning thing around that cross, but in fact, it keeps on training.
I have a i9-9900K with 64GB of ram so I dont think it is my computer. What is going on? I am simply asking the learner to STOP training, why is it so hard?
I am on 2020a.
  2 Comments
Sione Palu
Sione Palu on 2 Sep 2020
I ran a regression task using the ML Regression App, a few months ago on a 4 core machine (parallelized), with a dataset of about 12,000 unique rows and 33 columns/features and it took 4 days to run to completion. I'm gonna re-run the same task soon, but I'll probaby increase my AWS plan to 8 cores or more and I'm sure it's gonna speed up.
It's simply the GPR algorithm itself that is memory intensive. There's no way around it. I think it's best suited to running on a cluster.
Nelson Chow
Nelson Chow on 2 Sep 2020
Yeah, i get that GPR can be slow to run, but I just want it to stop.
At the moment, stopping gracefully is not possible. It takes hours for it to stop after hitting the cross.

Sign in to comment.

Answers (0)

Products


Release

R2020a

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!