0

# A question on Neural Network

I am an amateur programmer and have recently learnt how to build a simple neural network using python which has an output range of 0 to 1 using sigmoid activation function. But is there any way in which I can get an output value which is greater than 1? For instance how can I train my Neural network to predict the square root of a number?

1st Jun 2018, 2:20 PM
Vishal Krishna
6 Réponses
+ 4
Rational Agent Exactly. The activation function, whether a sigmoid, tan-h, relu or any other just indicates a "trigger" that activates (or not) a particular neuron. The shape of the function determines the likeliness of the activation especially in areas close to zero.
2nd Jun 2018, 6:41 AM
Kuba Siekierzyński
+ 3
Sigmoid function assumes 0-1 values only. It is a cumulative distribution function of probability density. And since probability cannot be higher than 1, sigmoid may not, either. By the way, the sigmoid function value is not the prediction of the value! It is just a probability that the given observation is of a predicted class.
1st Jun 2018, 4:19 PM
Kuba Siekierzyński
+ 1
why would you like to do that? Isn't it easier to use a function to calculate the square root of a number?
1st Jun 2018, 2:40 PM
Ulisses Cruz
+ 1
Just out of curiosity
1st Jun 2018, 2:40 PM
Vishal Krishna
0
Thanks. But isn't there any other way with which a Neural Network can be trained to output square root values.
1st Jun 2018, 4:38 PM
Vishal Krishna
0
their are different activation functions besides the sigmoid.however the sigmoid function isn't the networks output if I understand it righr
1st Jun 2018, 10:00 PM
Rational Agent
Aujourd'hui en vedette
XP farming