Choose a hidden activation function that makes sense and an output activation function that makes sense. The output activation function should map to the interesting domain. Look at the min and max Q values of Tabular vs. tabular and compare to neural vs. neural. Same for tabular vs. random and neural vs. random. Compare the distributions of Q values(?)
Read Efficient BackProp
Wiki on activation functions
Matlab Activation Functions
Choose a hidden activation function that makes sense and an output activation function that makes sense. The output activation function should map to the interesting domain. Look at the min and max Q values of Tabular vs. tabular and compare to neural vs. neural. Same for tabular vs. random and neural vs. random. Compare the distributions of Q values(?)
Read Efficient BackProp
Wiki on activation functions
Matlab Activation Functions