In #15782 (comment), @rkern writes regarding that fix:

FWIW, this broke a few downstream users where y_train.std() == 0 (e.g. only one datapoint) when being used for Bayesian optimization. This probably needs a guard for such a case (which is common in the Bayesian optimization use case)

More detail in bayesian-optimization/BayesianOptimization#243 (comment):

The normalize_y=True option which is used now divides out the standard deviation of the y data, not just subtracting the mean. When there is just one data point, this results in a NaN.