Detecting Training-Serving Skew
Find the normalization mistake that breaks this skew z-score check.
Codepython
def detect_skew(train_mean, train_std, serve_values):
serve_mean = sum(serve_values) / len(serve_values)
z = (serve_mean - train_mean) / len(serve_values)
return abs(z) > 3What is the bug in this skew-detection code?