C3_W1_Lab_2 - How would Example Validator work for unstructured data?

Course Q&A Machine Learning Engineering for Production(MLOps) Machine Learning Modeling Pipelines in Production
1

Hi Everyone, hope all has been well.

Up until now we had been working with structured datasets when working with Example Validator component, However, this is the first time we had used tfx for unstructured data, for Fmnist dataset, where the parameters are lmage and the label.

While there are no anomalies noted here in the exercise, I’m unable to comprehend how would Example Validator work for detecting anomaly for an unstructured data like Image/Text. I get that if there was an image with a different label since outside my domain, that would have gotten flagged but would Example Validator also look for semantic space for unstructured data and detect an anomaly? So if there was an image of a Bunny labelled as a T-shirt in my validation, would Example Validator detect this as anomaly?

2

Missing values and out of range of values should be minimal checks done for unstructured data. For image data, these should result in no anomalies.