Skip to content
Discussion options

You must be logged in to vote

Yes, burn supports this. Train and validation come from SupervisedTraining, and the test split goes through EvaluatorBuilder. The metrics are computed by the same metric types, so you get the same accuracy and loss calculation on all three splits.

The flow is:

  1. Pass the train and valid dataloaders to SupervisedTraining::new(...) and register metrics with .metrics((AccuracyMetric::new(), LossMetric::new())). Those are reported per epoch as Train and Valid in the TUI.
  2. launch returns a result that includes the trained model and the renderer.
  3. Build an EvaluatorBuilder with the same metrics, give it that renderer so the test results show up in the same TUI, and call eval (one test set) or eval…

Replies: 1 comment 1 reply

Comment options

You must be logged in to vote
1 reply
@hashcatHitman
Comment options

Answer selected by laggui
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Category
Q&A
Labels
None yet
2 participants