Add custom metric which keeps track of maximum of another metric during training

Hi all,

I’m tuning the hyperparameters of a Tensorflow model in Google AI platform, but I have the following problem: For the evaluation metric I want to optimize, it seems like only the metric value at the end of the training is reported to the hyperparameter optimizer, instead of the best value achieved during training. Is it possible to create a metric, which will track the maximum value over the entire training?