Professional Cloud Architect – Google Cloud Certification Guide

Hyperparameter tuning

When we were training our models, we made some assumptions regarding parameters. Those parameters are so-called hyperparameters. By tuning them, we can get higher accuracy rates. For example, we can use learning rate parameters, which indicate the extent of the changes made to W and b. If these parameters are too small, it can take a long time for our model to converge. If these parameters are too big, our model may never converge, thereby missing the ideal state. The other parameter we might want to tune is how many times we run through the training process. By changing it, we could potentially obtain higher accuracy.

Keep in mind that this process is rather experimental and requires some experience to get it set right. ...