Having problems with the trainnet function

13 views (last 30 days)
Andrew
Andrew on 27 May 2024
Edited: Jayanti on 14 Oct 2024
I am trying to fine tune a machine learning model. While using the trainnet function it states "Training stopped: Training loss is NaN" in the command window. The old trainNetwork functions works, which to me makes it weird that trainnet does not. How do I work around this? Possible solutions?
  1 Comment
Matt J
Matt J on 27 May 2024
Edited: Matt J on 27 May 2024
Please post enough of your code for us to run and reproduce the problem. Please also post the code as formatted text, e.g.,
imageSize=[256,384];
rather than as an image, so that it can be easily copy/pasted.

Sign in to comment.

Answers (1)

Jayanti
Jayanti on 14 Oct 2024
Edited: Jayanti on 14 Oct 2024
Hi Andrew,
I tried to run the code you provided on custom dataset, and it is working fine as you can see in the attached image.
Without actual data, it is difficult to provide a definitive solution upfront. However, the following two options can be explored to address this issue:
  1. Adjusting learning rate – If learning rate is very high it can cause the model to diverge, leading to NaN values. So, try reducing the learning rate.
  2. Gradient Clipping - Consider adding gradient clipping to prevent exploding gradients.
Hope it helps!

Categories

Find more on Image Data Workflows in Help Center and File Exchange

Products


Release

R2024a

Community Treasure Hunt

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

Start Hunting!