Multiple hyperparameter tuning

Now that you've successfully shown your ability to use hyperparameter tuning on the Banknote_Authenication dataset, it's time to explore whether adding another hyperparameter to your tuning run on the model will improve the outcome.

Note that, when using the canned dnn_classifier, the activation function defaults to relu. This method has the advantage of not throwing backpropogation errors and builds quickly compared to other models. Another activation function, softmax, gives a probability output and is best for recurrent networks and probabilistic models. Which is a better choice for the current model? Use tfruns to find out!

Question

If we create flags with dropout rates of 0.2, 0.3, 0.4 and two activation functions (relu and softmax), how many tuning runs will be undertaken?

Possible Answers

4 tuning runs
6 training runs
8 tuning runs

Exercise

Multiple hyperparameter tuning

Instructions 1/2

Question

Possible Answers