Hands-On Intelligent Agents with OpenAI Gym

Normalizing observations

In some cases, normalizing the observations can help with convergence speed. The most commonly used normalization process involves two steps:

1. Zero-centering using mean subtraction
2. Scaling using the standard deviation

In essence, the following is the normalization process:

(x - numpy.mean(x)) / numpy.std(x)

In the previous process, x is the observation. Note that other normalization processes are also used, depending on the range of the normalized value that is desired. For example, if we wanted the values after the normalization to lie between 0 and 1, we could use the following:

(x - numpy.min(x)) / (numpy.max(x) - numpy.min(x))

In the previous process, instead of subtracting the mean, we subtract the minimum value ...