model.add(Dense(20,input_dim=13,activation='relu'))
model.add(Dense(10,activation='relu'))
when we write code like this to add hidden layers in neural network then :
1. what this 20,10 values actually represent... are these neurons in hidden layers or units.. but units of what?
2.How we decide the above numbers.
3. Generally, many researchers take half the quantity (like 10 ) in the second hidden layer, why so?
1. Here you are right, the 10 , 20 stands for the number of nodes in the hidden layer. They are just the number and does not correspond to any physical quantity to have units. You can consider these layers as matrix operations. like 13,20 matrix with a relu function acting on a 20,10 matrix.
2. The numbers are proportional to the number of features you want to extract, or the number of useful feature you think there is, that will determine the out.
3. usually this reduction is used to narrow down on the bare minimum of features required for classification. You use 2/3rd of the value or even less. its not a standard to use exactly half. Just make sure the number of nodes decrease gradually.
You can contact me if any doubts exist:
In addition to the answer by Anoop, the number in the last layer will be the number of output neurons. The most important notice here is how to use activation function especially in output layer it depends on your task. For more details https://docs.w3cub.com/tensorflow~python/tf/keras/layers/dense/
Regards
I think, you are new in Deep Learning. If so, then you need to study this book. It will introduce you to the deep learning.
Book: Learn Keras for Deep Neural Networks
Author: Jojo Moolayil
- Badges
- Science topic