How Should you Architect Your Keras Neural Network: Hyperparameters (8.3)

[@BiancaAguglia]

This gave me an idea: it would be great to have an interactive dashboard (or a web app) that shows each hyper parameter as a widget. The user can play around with the values of the hyper parameters and actually see how the model and its results change. My github page is embarrassingly empty. I'll try my hand at building this dashboard. It would be the first project in my portfolio. 😊 Thank you for the inspiration, Jeff.

[@HeatonResearch]

That sounds pretty cool. Post a link if you add that!

[@BiancaAguglia]

@@HeatonResearch I will. Thank you for the encouragement. 😊

[@GeneralTerzX]

here you go playground.tensorflow.org/

[@joshuathomas2660]

@@GeneralTerzX Hello, Yesterday i have used this play exercises (NN) in my AI class. quite useful and interactive

[@hanserj169]

I was optimizing a binary classifier and It is working well with dropout=0.5 regarding accuracy and AUROC. Do you think this value is too high?

[@hhemsi6612]

I noticed in the method evaluate_network the second iteration creates a model with only one layer (Dense with activation=softmax) is this the intention? Thanks