-1
$\begingroup$

I am currently working with a dataset which has complex values. I need to normalize it such that I can use $\tanh$ as activation function, which is within $-1$ to $1$. But how can we normalize complex values?.. The normalizing should not remove the complex part, but also be normalized.

  • 1
    Explain more... – 2017-01-14
  • 0
    I am not sure what more there is to explain?.. – 2017-01-14
  • 0
    What's data-set volume? How many points has? How much points are big? – 2017-01-14
  • 0
    The dataset consist 1025 rows, but different number of columns. .. The dataset is stored as a matrix. All the points are complex values, and most of them are outside the range.. I just need one way of normalize it.. – 2017-01-14
  • 0
    the complex part should not be removed... but also be normalized. – 2017-01-14
  • 0
    Do you want to chat? – 2017-01-14
  • 0
    i cannot not enough rep – 2017-01-14
  • 1
    Let us [continue this discussion in chat](http://chat.stackexchange.com/rooms/51782/discussion-between-myglasses-and-user25778). – 2017-01-14

1 Answers 1

0

To clarify: The problem is that the output layer he wants to use has tanh as activation function and the complex numbers are the labels are complex numbers.

The solution is simple: Split the real and the imaginary part and use the neural network for regression on $\mathbb{R}^2$ instead of $\mathbb{C}$.

The normalization is simple:

  1. Subtract the mean value
  2. Divide by the range of the values
  3. Multiply with 2

However, if your range is not limited you should think about removing the tanh activation function.

You might also be interested in this answer to regression with neural networks.