features

click image to enlarge

User Interface for graphically building Bayesian networks & Dynamic Bayesian networks, which easily connects to a variety of data sources for setting evidence, Parameter learning, charting, Data Sampling, and performing Batch queries. View help file content

click image to enlarge

Dynamic Bayesian networks (DBN) for modelling time series or sequential data. View Time Series walkthrough

click image to enlarge

Data Explorer allows evidence to be loaded from a data source such as a spreadsheet or database, and transferred to a network, or charted. Supports discrete and continuous variables as well as Dynamic Bayesian networks (time series). View help file content

click image to enlarge

Query Explorer for displaying probability queries, such as the probability of variables, the history of variables and time series queries. Queries can be saved and restored at a later date. View help file content

• Calculate the probability of individual variables given evidence.  (e.g. P(A) and P(B) given the evidence.)
• Calculate the joint probability over multiple variables given evidence.  (e.g. P(A,B) given the evidence.)
• Calculate the probability of time series variables given evidence  (e.g. P(A_t=2) and P(B_t=5) given the evidence.)
• Calculate the joint probability of time series variables given evidence  (e.g. P(A_t=2 , A_t=5) given the evidence.)
• Calculate a range of time series queries given evidence  (e.g. P(A_t=1..25) given the evidence.)
• Calculate the log likelihood of evidence.

click image to enlarge

Batch queries for running predictions against multiple cases in a data source such as a database or spreasheet. Predictions can then be compared and charted. View help file content

Calculate the likelihood or log-likelihood of evidence set on a network. Both discrete and continuous variables are supported as well as Dynamic Bayesian networks (time series).

Log likelihood values are often used to detect anomalous data.

View help file content

click image to enlarge

As well as hard evidence, support is included for soft (virtual) evidence on both standard Bayesian networks and Dynamic Bayesian networks. Evidence can be entered via the Evidence window which allows Copying and Pasting from spreadsheets, or alternatively evidence can be set using Data Explorer. View help file content

As well as discrete variables, Bayes Server™ supports Continuous variables using Conditional Gaussian distributions. Support for continuous variables is also included for Dynamic Bayesian networks (time series).

click image to enlarge

Parameter learning, which supports multiple threads, both discrete and continuous variables, Dynamic Bayesian networks (e.g. Time Series), and learning with missing data. View help file content

The Bayes Server™ library (API), can easily be called by .NET languages such as C#, F#, VB.NET and C++.NET and many other languages that can interface to .NET libraries.

click image to enlarge

Data Sampling feature, allowing the generation of sample data to help visualize networks, and generate test data. Supports both discrete and continuous variables, Dynamic Bayesian networks (e.g. Time Series), and missing data. View help file content

Support for multiple variables per node. This allows, for example, the direct specification of mixtures of full covariance Gaussians.

Relevance optimization ensures only distributions relevant to a query are used.

Evidence propagation ensures implicit evidence is inferred from any explicit evidence. (e.g. Male=>Pregnant=False)

Support is included for disconnected networks.

Missing data support allows evidence on some variables to be omitted. As well as standard Bayesian networks, support is included for Dynamic Bayesian networks, nodes with multiple variables, and parameter learning.

Even though Bayes Server™ supports both discrete and continuous variables, sometimes it can be useful to discretize continuous data, generating a discrete variable, where each state represents a continuous interval.

click image to enlarge

Mesh queries allow visualization of predictions, by generating a 2-D surface plot. View help file content

click image to enlarge

A confusion matrix measures the performance of a Bayesian network when it is used for classification. View help file content

click image to enlarge

A lift chart measures the performance of a Bayesian network when it is used for classification. View help file content

Discrete and continuous histograms can be generated based on data loaded in data explorer, or generated by data sampling or batch queries.

Bayes Server supports queries which calculate the most probable configuration (explanation) of nodes/variables without evidence. For example, most probable explanation queries can calculate the most probable sequence in a time series model, using a generalized version of the viterbi algorithm.

Conflict is a measure that detects evidence that is conflicting or rare. The greater the conflict value above zero, the more likely the evidence is in conflict, or rare.

Nodes can be added to a network based on values in a data source such as a database or spreadsheet.

Retract evidence, is an option that allows predictions to be made on variables that have evidence assigned.

The direction of links can be reversed. This will maintain the overall probability distribution of the model, and may induce new links in order to do so.

An optional weight column can be included in a data source, allowing a support/probability (or any positive value) to be associated with a case. This is often used when a dataset contains large numbers of duplicate rows, or can be used to associate a probability to certain cases.