Your Own Blueprint
Build your own net from layers and neurons.
What you'll learn in this level
- What a net's blueprint is and how to design one yourself
- The difference between classification and regression
- Why not every blueprint fits every task
A Blueprint for a Net
So far you've looked at ready-made networks. Now you design one yourself. You decide how many hidden layers there are and how many neurons sit in each layer. That design is called the architecture.
Then you test your network on a task. There are two kinds: in a classification the network picks one of a fixed set of options, for example dog or cat. In a regression it estimates a number, for example how warm it will be tomorrow.
How you design a net
- 1
Set the input
The input layer gets its own neuron for each piece of information.
- 2
Choose hidden layers
You decide how many in-between layers your net gets.
- 3
Spread out the neurons
More neurons per layer means more room for patterns. But it also means more computing work.
- 4
Test and adjust
You try your net on a task. Then you change the blueprint until it runs well.
Guess first, then experiment
A blueprint worked great for a classification. Does it work just as well for a regression, so for predicting numbers?