Possible MES bug ?

[Shengyang Sun]

GPflowOpt/gpflowopt/acquisition/mes.py

Line 99 in cabaac0

 gamma = (fmean - tf.expand_dims(self.samples, axis=0)) / tf.sqrt(fvar) 

In the original paper, gamma is computed as (y* - mu) / sigma. It seems the code here is reversed.

[Nicolas K]

Hello!

The reason for this is that the paper is written in the sense that they want to maximize an objective, whiles our framework minimizes a cost function.

[Shengyang Sun]

ah, I see. Thanks