Generates data with drift
- Covariate distribution is initialised with a bayesian network.
- Covariate is drifted by linear interpolation of the conditional probability tables.
- Each unique covariate value combination is assigned a class via a tree with the following structure:
[x_1,x_2]
+
|
|
|
|
+----------------+
| | |
x_1=0| x_1=1| x_1=2|
v v v
y_1 | |
| |
+-----------------+ +-----------------+
| | | | | |
x_2=0| x_2=1| x_2=2| |x_2=0 |x_2=1 |x_2=2
v v v v v v
y_1 y_1 y_1 y_1 y_1 y_1
- Note that the first value of each attribute layer is always assigned a class
- The Posterior is drifted by changing the class assigned at the leaf nodes