C2_W2_finish Selecting the number of Units in each Layer of the NN

Course Q&A Machine Learning Specialization Advanced Learning Algorithms
1

image
image1546×524 108 KB

I just finished the lab of C2 W2. And throughout the week I was wondering how the number of units in each layer gets selected?

I just rewatched the Demand Prediction Video where the architecture of neural network is described, choosing the number of layers and number of units/neurons within each layer. If the power of the neural network is that it will basically feature engineer on it’s own and then determine which of those features are relevant, how does one determine the structure that happens in?

The video mentions we will learn more about NN architecture later on, could anyone share when in the course it happens? =] so I can be extra excited for that video.

2

It’s by experimentation.

The overall goal is to get a model that works well enough, without being so complicated that training it is difficult or takes too many resources (time, memory, processing power, etc).

The NN just learns the weights that give the lowest cost. If a weight value is small, that means the feature doesn’t have much effect on the results.