C3W1 Lab - accuracy for cirrus sample


On the AutoML task, the cumulonimbus example is correctly classified with 98% accuracy, but the other example also classifies as cumulonimbus, with only 57% accuracy, which does not satisfy the requirements of the last part of the lab.

Not sure what I could possibly tweak given that it’s AutoML? And that we have a fixed sample of training data?

For training + serving took about an hour, so not quick to repeat.


Hi Scott! Normally, you will feed more data to improve the model’s performance. You can also check the images if some are mislabeled. Re: the grader, there is a recent issue with the final progress check. You should receive a perfect score with the default settings provided in the instructions and you don’t have to tweak and re-label the images. I have submitted a report to the Qwiklabs team and they have acknowledged that there is indeed an issue that they will need to investigate. Please follow this thread for updates regarding this. Thank you!