Algorithm 2 DL training procedure.

Input: Data set $\Lambda$, training data number $n_{train}$, training batch size B, total training epoch number ${\Gamma }_{max}$ and learning rate $lr$.

Output: A neural network $\widehat{Q}$ with $\widehat{\omega }$.

1: initialization: Set the first training epoch number $\Gamma =1$ and a neural network Q with random weight $\omega$.

2: Randomly select $n_{train}$ data from $\Lambda$ as ${\Lambda }_{train}$ and the rest to testing data set ${\Lambda }_{test}$.

3: while $\Gamma <{\Gamma }_{max}$ do

4:  Randomly select B data from ${\Lambda }_{train}$, and compute the sum of the cross entropy loss.

5:  Compute the gradients of $\omega$.

6:  Perform Adaptive Moment Estimation (Adam) gradient descent training to update $\omega$ using the obtained gradients.

7:  Calculate the accuracy of neural networks replicating human behavior in training data set ${\Lambda }_{train}$.

8:  Perform $\Gamma =\Gamma +1$

9: end while

10: Calculate the accuracy of neural networks replicating human behavior in training data set ${\Lambda }_{test}$.

11: return the trained neural network $\widehat{Q}$ with $\widehat{\omega }$.