# How to Set Dynamic Customer Classes¶

Ciw allows customers to probabilistically change their class after service. That is after service at node k a customer of class i will become class j with probability $$P(J=j \; | \; I=i, K=k)$$. These probabilities are input into the system through the Class_change_matrices keyword.

Consider a one node system with three classes of customer. After service (at Node 1) customers always change customer class, equally likely between the two other customer classes. The Class_change_matrices for this system are shown below:

$\begin{split}\begin{pmatrix} 0.0 & 0.5 & 0.5 \\ 0.5 & 0.0 & 0.5 \\ 0.5 & 0.5 & 0.0 \\ \end{pmatrix}\end{split}$

This is input into the simulation model by including Class_change_matrices keyword when creating a Network object:

>>> import ciw
>>> N = ciw.create_network(
...     Arrival_distributions={'Class 0': [['Exponential', 5]],
...                            'Class 1': ['NoArrivals'],
...                            'Class 2': ['NoArrivals']},
...     Service_distributions={'Class 0': [['Exponential', 10]],
...                            'Class 1': [['Exponential', 10]],
...                            'Class 2': [['Exponential', 10]]},
...     Transition_matrices={'Class 0': [[1.0]],
...                          'Class 1': [[1.0]],
...                          'Class 2': [[1.0]]},
...     Class_change_matrices={'Node 1': [[0.0, 0.5, 0.5],
...                                       [0.5, 0.0, 0.5],
...                                       [0.5, 0.5, 0.0]]},
...     Number_of_servers=[1]
... )


Notice in this network only arrivals from Class 0 customer occur. Running this system, we’ll see that the count of the number of records with customer classes 1 and 2 more than zero, as some Class 0 customers have changed class after service:

>>> from collections import Counter
>>> ciw.seed(1)
>>> Q = ciw.Simulation(N)
>>> Q.simulate_until_max_time(50.0)
>>> recs = Q.get_all_records()
>>> Counter([r.customer_class for r in recs])
Counter({0: 255, 2: 125, 1: 105})


Note that when more than one node is used, each node requires a class change matrix. This means than difference class change matrices can be used for each node.