Artificial Intelligence By Example

Numerical –  controlled convergence

This approach can prove time-saving by using the target result, provided it exists beforehand. Training the reinforcement program in this manner validates the process.

In the following source code, an intuitive cross-entropy function is introduced (see Chapter 9, Getting Your Neurons to Work, for more on cross-entropy).

Cross-entropy refers to energy. The main concepts are as follows:

• Energy represents the difference between one distribution and another
• It is what makes the system continue to train
• When a lot of training needs to be done, there is a high level of energy
• When the training reaches the end of its cycles, the level of energy is low
• In the following code, cross-entropy value (CEV) measures ...