Question: Day4: Hyperparameter tuning practice: Error metric changes without any changes to hyperparameters

I am trying to execute the block just above the exercise 1. MSE value changes for each execution. Ideally it should not else it is impossible to tune hyperparameters. I confirm I have executed all the cell above this section, including seed. Is there something wrong, cant figure it out. Please do check. Thanks!

Hi @elugan
Yes you are right! The MSE will change for each execution because of randomness in neural networks. You need to use seed to stop the randomness (it’s shown in the linear regression notebook). Also, note that seed will not work in colab notebook but if you execute locally it should work.

Cant use locally due to tensorflow error on my system. Hyperparameter tuning is a very important thing I suppose, if colab cant handle the seed then it sounds like a hard stop on tuning part for me.

Hi @elugan
It is normal to obtain slightly different results after training Neural Networks since model parameters (weights) are randomly initialized at the start of training.

When optimised well, the differences due to the randomness won’t even matter much.

There’s nothing to worry about.