|
|
|
How do we learn such models
from a database. If an expert
specifies the dependency graph of the PRM, the task of estimating
|
|
the parameters is fairly
easy.
|
|
E.g. to estimate
P(A|high,easy) we simply compute the ratio of A-grades received by highly
intelligent students in easy classes
|
|
To all grades received by
highly intelligent students in easy classes.
|
|
Note that we can estimate
each cond prob table independently of others.
|
|
|
|
However, often variables of
interest are not all observed in the database. How can we estimate parameters in such
cases?
|