Trigger Word Detection modelf function Dropout Rate


In the Trigger Word Detection homework, the instructions above the modelf function state to use a Dropout layer with a keep rate of 0.8 in a couple different places, for example.

“Follow this with dropout, using a keep rate of 0.8”

In the keras documentation, it seems that the Dropout layer is defined where the “rate” hyperparameter is actually the fraction of units to drop (i.e. this value is the drop rate not the keep rate).

If you take this into account and specify the layer with rate=0.2, then the test cases will throw an error. If you put 0.8 like the instructions indicate, the test cases pass.

Not a big issue, but it seems like what we are actually building is a Dropout layer where 80% of the inputs get turned off. Could be an easy fix by either changing the instructions to 0.8 drop rate or by changing the test cases to look for 0.2.

Please let me know if it turns out I am actually just misunderstanding something here.

Thank you,

1 Like

Thanks for your observation. I’ll look into it.