Train on a Dataset

Required Steps

• Once you are satisfied with your dataset, you can send the dataset to the training pipeline to create a model by clicking on the "Train" button. Training a model can take some time. How long can vary wildly depending on the amount of input data and on the model type. For a classic classifier, it is a safe bet to assume you will wait a minimum of 20 minutes. For a few shot classifier, the average minimum time is 10 minutes.

• After training has finished, you can see the test accuracy. Click on the Details button to get more information about your training. Under the tab Training Confusion Matrix and Training Summary, you will see some information on the training of your model and how accurate the results might be. You can see everything from an overall accuracy rating of your data (as derived from testing the model against some test images from your collection for which we already know the correct answer), down to a label-by-label interactive graph with detailed information on each item.

• If you have trained a Few Shot Classifier, you might not have had test sources, and hence no test accuracy is available. We recommend to test the performance of your model by packaging it (Step 5) and testing it in the Data App.

If at any point you need help from the Passio team, please reach out to us at [e-mail]