Menu: Network menu

From DSL

The Network menu allows for performing operations that relate to the entire network. It offers the following commands:

• Network Properties

The Network properties invokes the Network Property Sheet for the current model. The network property sheet can also be invoked by double-clicking in any clear area on the main model Graph view window. See Network Property sheet section for more information.

• Update Beliefs ( Shortcut : F5 )

The Update command runs the selected algorithm on the model and brings the values of each of the variables of interest up to date. The Update command can be also executed by pressing the Update tool from the Standard toolbar. The algorithm to be applied can be selected from this menu i.e. the Network menu. For more information on various Inference Algorithms supported by GeNIe, see Applying Inference Algorithms section.

• Update Immediately ( Shortcut : Ctrl+F5 )

The Update immediately command toggles between two modes of operation: eager and lazy updating of models. When the Update immediately flag is set (default), every time the model is modified or every time evidence is entered, an updating algorithm is executed to make all values up to date and valid. When the Update immediately flag is cleared, it is the user's responsibility to update the model (see Update Beliefs command above) when values are desired for examination.

• Invalidate values

Normally, GeNIe will take care that all values in the nodes that are not marked as Invalid are up to date. Occasionally, because of an (unlikely ) error in the program, the values in nodes that are not Invalid may be wrong. Also, if you have run a stochastic sampling algorithm and would like to recompute the values with the new number of samples (larger number of samples give you a higher precision), you will need to invalidate all values and force GeNIe to recompute them. The Invalidate values command is a manual escape for such situations. It will invalidate all values in the network and allow to update them, which in most cases should fix the problem.

• Clear all targets

The Clear all targets command reverses the effect of any Set target command executed prior to Clear all targets. All (Target) status icons will disappear.

• Clear all evidence

The Clear all evidence command un-observes all nodes in the network, i.e., it reverses the effect of any Set evidence command executed prior to Clear all evidence. All (Observed) status icons will disappear.

• Clear all decisions

The Clear all decisions command parallels the Clear all evidence command, but it acts only on decision nodes. It clears all decisions in the network, i.e., it reverses the effect of any Set decision command executed prior to Clear all decisions. All (Observed) status icons will disappear from the decision nodes.

• Enable Temporal Plate:

This command enables a temporal plate which can be used for creating and performing analysis on Dynamic Bayesian Networks.

• Slice Count:

This is a supporting feature of DBN that allows the user to select a slice count for temporal analysis.

• Unroll:

This is a supporting feature of DBN that allows unrolling a temporal network for a given number of time-slices.

• Probability of Evidence:

This command displays the evidences, probability of recent evidence and the log of it in a new window. If there are no evidences, no probabilities will be displayed.

• Algorithm

See Inference Algorithms section under Elements of GeNIe for more detailed information on each algorithm.

• Annealed MAP

The Annealed MAP algorithm solves the problem of finding the most likely configuration (MAP) of a set of nodes given the evidence on another set of nodes. The result of this algorithm is an approximate solution. However, this algorithm drastically extends the class of MAP problems that can be solved.

• Value Of Information (VOI)

Value of Information is used to calculate the expected value of information in the network for the specified decision node. Value of Information calculates only Expected Value of Perfect Information. Decision nodes have to be present in the network for VOI to be calculated. See Tutorial 11: Value of Information to learn how to use VOI.

• Learn Parameters

The learn parameters command is used to make an existing network learn a new set of data. See Tutorial 16: Parameters to learn about this feature.

• Generate Data File

The Generate Data file command is used generate a text file, containing records that is representative of the network. While the records lists the different states of the model, it can also be used to learn networks.

• Obfuscate

The Obfuscate Network command can be used to create a new network from an existing network with an option to change various model properties while maintaining the original structure. This kind of networks is very useful when the you need to share the models with other groups.

• Strength of Influence

Displays colored lines varying in width based on the strength of influence of the nodes. Enable Diagnosis

• Enable Diagnosis

The Enable Diagnosis checkbox is used to enable/disable diagnostic features of GeNIe.

• Bayesian network algorithm selection

It is up to GeNIe user to select an algorithm that is appropriate for the model at hand. The Network menu allows for choice of the default Bayesian network algorithm (marked with a checkmark). Whenever updating takes place, the current default algorithm will be executed. The clustering algorithm, which is GeNIe default algorithm, is the fastest known exact algorithm and it should be appropriate for most models. Only when models become very large or very densely connected, the clustering algorithm may turn out to be too slow. In this case we suggest selection of one of the sampling algorithms. See Inference Algorithms section of Elements of GeNIe for more detailed information on each algorithm.

• Influence diagram algorithm selection

Similarly to the selection of a Bayesian network algorithm, it is up to GeNIe user to select an algorithm that is appropriate for the model at hand. The Network menu allows for choice of the default influence diagram algorithm (marked with a checkmark). Whenever updating takes place, the current default algorithm will be executed. We would like to note that all influence diagram algorithms are based on the Bayesian network algorithms and the choice of a Bayesian network algorithms will have impact on the precision and performance of the influence diagram algorithms.