Netica

Netica is a powerful, easy-to-use, complete program for working with belief networksand influence diagrams. It has an intuitive and smooth user interface for drawing the networks, and the relationships between variables may be entered as individual probabilities, in the form of equations, or learned from data files (which may be in ordinary tab-delimited form and have "missing data").

Once a network is created, the knowledge it contains can be transferred to other networks by cutting and pasting, or saved in modular form by creating a library of nodes with disconnected links. Of course, the networks and libraries may be saved in files or printed out.

Netica can use the networks to perform various kinds of inference using the fastest and most modern algorithms. Given a new case of which we have limited knowledge, Netica will find the appropriate values or probabilities for all the unknown variables. These values or probabilities may be displayed in a number of different ways, including bar graphs and meters. The case may conveniently be saved to a file, and later brought back into the network (or a different network) for further querying, or to take into account new information about the case. Netica can use influence diagrams to find optimal decisions which maximize the expected values of specified variables. Netica can construct conditional plans, since decisions in the future can depend on observations yet to be made, and the timings and inter-relationships between decisions are considered.

Netica can be used to transform a network in a number of ways. Variables that are no longer of interest may be removed without changing the overall relationships between the remaining variables (technically, the probabilities are "summed out" when we don't know the variable's value, and a more complex operation is used when we do). Probabilistic models may be explored by such operations as reversing individual links of the network, removing or adding causal influences, optimizing one decision at time, etc. These operations may be done with just a click of the mouse, which makes Netica very suitable for easy exploring, and for teaching belief network and influence diagram concepts.

• Generate websites from Bayes nets. Example site
• Generates presentation quality graphics which can be incorporated into other documents.
• Version 5.0 and later can do TAN learning of link structure from data.
• Compiles belief (Bayesian) networks into a junction tree of cliques for fast probabilistic reasoning.
• Extensive built-in and online help.
• Utility-free sensitivity analysis.
• Can generate highly customizable reports on many aspects of the Bayes net, nodes, states, CPTs, cases, findings, beliefs, sensitivity results, other inference results, etc.
• Can test the performance of a network using a file of cases.
  Netica will print out a confusion matrix, error rate, logarithmic and quadratic (Brier) scoring rule results, calibration table and surprise indexes for each node desired.
• Can find optimal decisions for sequential decision problems (i.e., later decisions are dependent on the results of earlier ones).
• Can solve influence diagrams efficiently by using clique trees.
• Can learn probabilistic relations from data, even with missing data.
• Provides easy graphical editing of belief networks and influence diagrams, including:
• cut / paste / duplicate nodes without losing their probabilistic relation
• many ways of displaying the nodes (bar graphs, meters, etc.)
• links with bends to keep complex diagrams orderly
• allows comments, keeps track of author, when changed, etc. for each node
• unlimited levels of undo / redo
• Allows the entry of probabilistic relations by equation, with an extensive built-in library of probabilistic functions and other mathematical functions.
• Has facilities for the easy discretization of continuous variables.
• Can reverse individual links and "sum out" nodes of influence diagrams or belief nets, for model exploration.
• Supports disconnected links, which makes possible libraries of probabilistic relationships (See Figure 1).
• It is possible to represent networks with nodes whose values change over time (a persistence is defined for such nodes), and to have links with time delays (which allows cycles). The software can automatically convert these networks into expanded regular networks covering a limited period of time (See Figure 2).
• Accepts likelihood findings (i.e., virtual evidence), and findings of the form that some variable is not in some state.
• Can display nodes and inference results in a number of forms, including bar graphs of beliefs and a true/false meter (See Figure 3).
• Easy to customize the display to be suitable for an end-user.
• Supports documentation and tracking of every node and network (with comments, titles, author, when last changed, etc.)
• Has no built in limits on the size or complexity of networks, so they are limited only by available memory.
• Can work hand-in-hand with the Netica API product (for example, sharing the same files).