Weka Multiple Chuck with Multiple Hidden Layers

Question

I am trying to use Multi-Perceptron in the Weka knowledge flow. In the application you can see the settings for the block. As written in the help: "hiddenLayers" - defines the hidden levels of the neural network. This is a list of positive integers. 1 for each hidden layer. Comma seperated. To avoid hidden layers, hover over here 0. Roll it over only if autobuild, as well as the substitution values ​​"a" = (attributes + classes) / 2, "i" = attributes, "o" = classes, "t" = attributes + classes. "

However, this is still a bit confusing to me. How can I build a neural network with three hidden layers, each of which has a different number of units (say, 2,5,6). And can you explain the meanings of wildcards? I think this is only for the number of hidden layers, and not for the number of hidden units in each layer.

Thank.

Answer 1

The GUI option will help you with this. If you add “2,5,6” to the hidden layers, it will create 3 layers with 2, 5 and 6 units respectively. Wildcard values ​​appear to be shortcuts to the numbers they represent (a = (# from attributes + # classes) / 2, etc.). Here are a couple of visual steps that represent what I mean.

---

You can see "a, 2, 5, 6" in hiddenLayers.

a = # + # /2

6 1 , a = 7/2 = 3

, 3 , 2, 5 6, 1 .