Home>

I am always studying.

I'm trying Deep Q learning with Chainerrl.
Tell me about data standardization.

If a continuous variable takes a value between -10 or less and 10 or more across 0 as shown in the following list:
How should we standardize it?
Since we are going to use it with nominal variables separated by (0, 1), we want to make the values ​​of other continuous variables also in the range of 0 to 1. What should I do?
Although it seems that the values ​​are almost in the range of -10 to 10, the exact value is the minimum and maximum values.
Should I normalize -10 or less as -10 and 10 or more as 10?

[-0.0023340981421771097,
 10.10886891470353441,
 -0.8278840168666244,
 0.0,
 -10.17769576921076302,
 0.0,
 0.0,
 0.0,
 0.0,
 -1.4023147212763036,
 0.0,
 0.0,
 0.0]
  • Answer # 1

    Mathematically, if you want to change from -10 to 10 from 0 to 1, you will have to translate by a factor of 20 times from -0.5 to 0.5 and then the whole translation by 0.5.

    I just feel that it has a standardized function.

    If you don't know the maximum and minimum values, you can find each data.