Understand your Model's Performance

In the performance tab, you can see by class performance on the training, test and validation set. To visually inspect the predictions of the model, we also surface a set of best and worst predictions as well as false positives false negatives which can be filtered by class.


Performance Overview Tab

You can take a deeper look at all the data samples and advanced performance metrics using the link to Scale's Nucleus product.
If you think performance is looking great, you can move forward to download or deploy your model. If not, you might want to re-train the model with more data, which will create a new Model Version. All model versions are listed in the version history tab. Clicking into each version shows the same details for your most recent model. The table also indicates which is the latest model version as well as the model version used in the production endpoint.

Updated a year ago