The key assumption in machine learning is that your training dataset and the real world data have similar distributions. But what happens when this is not the case?
Unidentified data shifts result in poor model performance and delays in deliverables as a result of misdirected model tuning approaches.
In this session we will discuss the importance and methods of identifying shifts in your datasets as well as break down the different types of shifts and their effect on model accuracy. We will explore practical solutions of handling machine learning problems with shifted data in order to achieve more robust and production ready models with higher reliability.