Is this from a quiz, or just a general question?
If the latter, it’s based on your experience and experimentation. The goals are to get good enough results while minimizing the computational cost of training.
The latter. Thanks! But it there some general rule of thumb to select the initial number of layers and nodes? Will I learn that in week 4? If yes, sorry if I am jumping ahead.
Nothing concrete. Just guidance like this:
- Start with one hidden layer.
- Adjust the number of units to see if you get “good enough” performance.
- If you don’t get good enough performance, considering adding another hidden layer.