Kinds of Tables

The table selector allows choosing between the different types of tables that a node can have: a function table, conditional probability table (viewed with decimal or percentage numbers), experience table, unnormalized probability table and a counts table.

Can't Set:  The only type of table that a deterministic node can have is a function table, so if the deterministic selector to the left of the table selector is set to "Deterministic", any choice of the table selector other than "Function" will just result in a beep and an explanatory message going to the Messages window.  If the left selector is set to "Chance" instead, then any choice from the right selector is okay, except "Function", which is then the one that results in a beep.

Function Table:  The function table allows only a single output value for each possible set of parent values (i.e. a mathematical "function").  If the node is discrete, then there are a finite number of choices and Netica lets you enter them by clicking on a table cell and then choosing from a popup menu.  If the node is continuous, not discretized (such as a utility node), then Netica lets you enter a real number in each cell.

Conditional Probability Table (CPT):  These are the most common tables to work with, and are described in detail on the other pages of this chapter.  They provide a probability for each state of the node, given the condition specified by the row (i.e. each parent node having some value), so the probabilities of each row must sum to one.  You can enter them as decimal fractions or percentages, depending on whether the table selector is set to Probability or % Probability.

Experience Table:  Experience tables provide one confidence number for each row of the table.  Experience numbers must not be negative, and an experience of zero corresponds to an impossible condition (a row of Xs in the CPT).  Experience numbers mean approximately the equivalent number of matching cases seen.  Experience numbers are generated as the normalization factors of the "Unnormalized table", which in turn is the Counts table with a constant base experience added to each cell (usually 1).

Counts Table:  These are the actual counts of matches during the learning process, so they are generally integers (e.g. there were 7 instances matching this table cell).  However, they will have a fractional part if the case file had a "NumCases" column containing fractions, a learning degree other than 1 was used, the table was hardened or softened, a more advanced learning algorithm like EM or gradient descent was used or a fractional starting experience was entered before learning.