Question

Briefly explain training-serving skew and how it is addressed.

Answer 1

Training -Serving skew

Training serving skew is the common problem encountered when running machine learning model in production

Where the data seen at serving time differs in some way from the data used to train the model,which leads to reduced prediction quality /reduces performance

The common reason for the issue to arise are discrepancy in how data is handled in training compared to serving ,that is change in between serving and training

One of the best solution to addressing the training-serving skew is monitoring the system closely

And some other to address them are taking the same code from the historical data during training and reuse it during predictions

By measuring skew we can avoid the training serving skew that is by measuring the difference between training and serving data and their performance