Nature node:  A nature node in a Bayes net represents some variable of interest.  It may also appear in a decision net in which case it is a variable that cannot be directly controlled by the decision maker (i.e. it is determined by nature).  If a nature node has a functional relationship with its parents, it is called a deterministic node, whereas if the relationship is probabilistic, it is called a chance node.  The characteristic shape for a nature node is an ellipse, or a rectangle with rounded corners.

Negative finding:  A negative finding is a finding that some node is definitely not in some particular state.  Compare with positive finding and likelihood finding.  more info

Net:  In Netica documentation, the word net is used to mean a Bayes net or a decision net.

Netica:  Netica is a program created by Norsys for working with Bayes nets and decision nets.  more info

Netica API:  Netica API (also known as “Netica Programmer’s Library”) is software that you can link with your own programs to achieve much of the functionality of Netica Application.  It is created by Norsys and is designed for working with Bayes nets and decision nets.  more info

Netica Application:  Netica Application is the Netica product with an easy-to-use graphical interface for building and working with Bayes nets and decision nets.  To program Netica, use Netica API instead.

Netica-Web:  See AutoNetica.

No-forgetting links:  If a decision maker remembers the decisions he made at an earlier time, and also the knowledge he had available to him at that time, then in his decision net there will be informational links going from earlier decision nodes and their parents, to later decision nodes.  These are called no-forgetting links.  more info

Node:  A node is a component of a Bayes net or decision net used to represent a variable (i.e. scalar quantity) of interest, and in Netica is usually drawn as a rectangle, rounded rectangle, circle or flattened hexagon.  more info

Node dialog box:  To change or view the properties of a node, such as its name or the states it has, you use a node dialog box, which you obtain by double-clicking on the node, or selecting it and then pressing the enter key.  To change its relation with its parent nodes, you use a table dialog box.  more info

Current state:  The current state of a node dialog box, is the state displayed next to the label “State:”   It may be changed by using the popup menu next to the “States:” label.  The state interval thresholds or state value displayed is for the current state.  more info

node name:  The node name text edit box in the node dialog box looks like this.

node title:  The node title text box in the node dialog box looks like this..                                   

states label:  The states label in the node dialog box looks like this.                   

Node relationship:  A node relationship, or node relation for short, is the relationship between a node and its parents.  It may provide the value of the node as a function of its parents’ values, or it may provide a probability distribution for the node depending on its parents’ values.  It is often expressed as a CPT in which case it can be viewed or edited using the table dialog box.  Alternately, it may be expressed as a probabilistic or deterministic equation.

Non-extreme probability:  A non-extreme probability distribution (also known as “strictly positive”) is one where the probability is never 0.  That means that it is also never 1, and that it has no points of complete “certainty”.

Normal distribution:  The normal distribution (also known as “Gaussian distribution), is the most commonly used continuous distribution with infinite support.  more info

Normalize:  Menu item, Table → Normalize.  Not supported after version 3.08.

Normative theory:  A normative theory does not indicate what agents usually do (which is a descriptive theory) or what agents ought to do (which is a prescriptive theory), but what agents must do if they wish to act optimally in a given situation, where optimally is defined in a particular way with respect to the situation.

Norsys:  Norsys Software Corp. is the company which develops Netica Application and Netica API.  You can get more information about Norsys from their web site at: www.norsys.com.  Questions and comments are very welcome, and may be sent by e-mail.

Optimal policy:  The optimal policy (also known as the set of optimal decisions) is the policy which results in the greatest expected value for the sum of the utility nodes (or one of those policies if there are more than one which result in the same expected utility).  Finding the optimal policy is sometimes called “solving” a decision net.

Outcome:  The outcome is the result of an event, or series of events, that could have turned out in one of several ways.

Parameter learning:  Parameter learning is the automatic learning of the specific relationships nodes have with their parents using case data, once it has already been determined which nodes are the parents of each node.  These relationships are usually in the form of conditional probabilities, or the parameters of a conditional probability equation.  Compare with structure learning.  more info

Parent node:  If there is a link going from node A to node B, then A is said to be a parent node of B.  Some people refer to it as a “direct predecessor”.

Path:  A path is a sequence of nodes from a net, such that you can get from one node of the sequence to the next node by traversing a link between them (but not necessarily in the direction of the arrow).  Compare with directed path.

Poisson process:  A Poisson process is one in which events occur randomly and independent of each other.  The number of events that occur in a fixed time period is given by the Poisson distribution, the time between successive events is given by the exponential distribution, and the time required for the occurrence of a fixed number of events is given by the gamma distribution.

Policy:  A policy (also known as a “control law”) is a set of decision rules, with one for each decision node of a decision net.  When Netica “optimizes decisions” it finds the policy which maximizes the expected value of utility.

Positive finding:  A positive finding is a finding that some node is definitely in some particular state.  Compare with negative finding and likelihood finding.  more info

Probabilistic inference:  Probabilistic inference is the process of calculating new beliefs for a set of variables, given some findings.  Technically speaking, it is the process of finding a posterior distribution, given a prior distribution, a model and some observations.  Bayes nets do probabilistic inference by belief updating.

Probability density function:  The probability density function (also known as “pdf”), is a function that provides the probability of a continuous probability distribution at each point within the distribution.  It may be integrated over a region to determine the probability of that region.  It is nowhere negative and its integral over the whole distribution is always 1.  The integral of the pdf from negative infinity to x is known as the cumulative density function (cdf).

Probability revision:  Probability revision is the process of adjusting the conditional probability tables of a Bayes net to account for a new case (i.e. set of findings), or more often, for a new set of cases.  It is a form of parameter learning, which generally involves learning from cases.  Compare with belief updating.

Prospect:  A prospect is the probability distribution over possible outcomes, given a policy and some findings.

qNetica:  See AutoNetica.

Query node:  See Target Node.

Relation symbol:  The relation symbol is a green tree structure which looks like this:  It is used on toolbar buttons to indicate the relation a node has with its parents.  For example, the button is to view or edit a table, the  button removes a node’s table, and the  button with a dice randomizes a table.

Reports:  Netica can generate a multitude of text reports, useful in understanding the information in your net.  These reports include:

Report → Beliefs, which lists the current beliefs (i.e. posterior probabilities) for nature nodes, and expected utilities for decision nodes.  example

Report → CPT Tables, which lists the node relation as a conditional probability table (CPT) or function table.  example

Report → Elimination, which lists the order used during compiling.  example

Report → Equations, which lists all equations of nodes.  Choosing the Horizontal Format option prints them in internal form.  example

Report → Excel, which hot-links to the node beliefs.  example

Report → Findings, which lists the findings (i.e. "case" or "evidence") currently entered, including likelihood ("virtual") findings.  example

Report → Junction Tree,  gives details of the net compilation process.  example

Report → List Selected, generates a list of the names of the nodes currently selected.  example

Report → Node Sets, which lists the nodes within the requested set.  example

Report → Network, which gives a summary information on the whole net.  example

Retracted:  Anytime after a finding has been entered into a node, that finding may be removed, or retracted.  After doing belief updating for the net, it will be as if the finding had never been entered.  more info

Right-click:  To right-click on something, place the mouse pointer ("cursor") over it, then press the right mouse button and choose an item from the menu that comes up.   more info

Root node:  A root node is a node with no parents.  See also leaf node.