Continuous-Discrete Node Value Transfer Function
The Continuous-Discrete node Value Transfer function
(CDVT function) is available in Analysis Menu under the Network Menu.
In essence the CDVT function implements a modelling
technique to circumvent the limitation of no discrete
children of continuous nodes in CLG Bayesian networks. A
discrete interval node is linked to a continuous node such
that the expression generating the distribution of the
discrete node is equal to the posterior marginal of the
parent continuous node. In this way the discrete node
becomes a discretize representation of the continuous
node. The discrete node has no table as the table is
regenerated using the Table Generator functionality. The
states of the discrete node are determined by the user prior
to compiling the network.
|
Figure 1: The dashed arrow specifies a CVDT function link. |
Figure 1 shows a link between a continuous and discrete nodes pair created by the CDVT function.
|
Figure 2: The Discrete Score is a discrete approximation of the Continuous Score computed using Table Generator functionality. |
Figure 2 illustrates how the density function of a
continuous node can be approximated by a discrete
distribution using the CDVT function.
The CDVT function has some major limitations such as
evidence can only be propagated along the direction of the
CDVT link, only sum-normal equilibrium is supported and no
type of analysis is supported on a network with CDVT links.
The discretization link should be used with extreme care.
Back