Version Beta - Sunday, July 8, 2018

  • .NET API now targets .NET Standard 2.0, with support for .NET 4.6.1+, .NET Core 2+ and more
  • New JavaScript & TypeScript API
  • Retracted Analysis for anomaly detection diagnostics
  • Fix for 'Add nodes from data' with decimal data type (e.g. currency).

Version - Wednesday, September 20, 2017

  • Fix for MySql .NET connector in User Interface.
  • Fix in user interface in parameter learning wizard with noisy nodes

Version - Thursday, July 20, 2017

  • Performance fixes to Impact Analysis and Log-Likelihood analysis with larger numbers of variables.

Version - Wednesday, July 19, 2017

  • Log-likelihood analysis tool.
  • In-sample anomaly detection. Helps remove anomalous data from a training set.
  • API: DataTable default implementation of IEvidenceReaderCommandFactory. Helps create partitioned data sets more easily.
  • API: kFold cross validation now includes non-callback version.

Version - Saturday, July 15, 2017

  • Link strengths can now be displayed graphically in the network viewer
  • Layout in compact mode now takes into account label bounds
  • Nodes can now be added to groups via selection in the network viewer
  • Changes made to Compact/Query node views now support undo/redo

Version - Tuesday, July 11, 2017

  • Node groups. Groups can be colored for display in the User Interface
  • New compact node view in user interface
  • New icons in User Interface
  • Network viewer magnifier

Version - Tuesday, July 4, 2017

  • New Hierarchical structural learning algorithm. Builds layered groups of variables.
  • New structural learning wizard.
  • New parameter learning wizard.
  • Add nodes from data (VariableGenerator) now reports % of missing data for each column.
  • Batch query User Interface no longer keeps handle open to data connection after processing.
  • Structural learning User Interface now supports learning with a subset of nodes.
  • Auto Insight User Interface now supports filtering test variables.
  • Search/Score structural learning now supports link constraints.

Version - Saturday, June 10, 2017

  • Discretized ranges now displayed in UI elements
  • Bug fix for continuous <-> discrete nodes in Chow-Liu and TAN structural learning algorithms
  • Link associations: fixed incorrect listing of variables and nodes in User Interface

Version - Tuesday, May 23, 2017

  • Impact analysis.

Version - Saturday, May 20, 2017

  • Auto-insight now supports continuous test variables in addition to discrete.
  • Auto-insight now reports the Kullback-Leibler divergence for discrete & continuous variables.
  • API now supports KL Divergence for discrete or continuous query distributions.

Version - Thursday, May 18, 2017

  • Value of information now supports discrete, continuous & mixed/hybrid networks.
  • New Link strength / association calculator
  • Value of information for decision graphs
  • Mutual information calculator now supports mixed discrete-continuous distributions
  • Entropy calculator now supports mixed discrete-continuous distributions
  • Kullback-Leibler divergence calculator.
  • Feature importance (selection) now includes mutual information.
  • Auto-insight now supports copy to clipboard.

Version - Thursday, May 11, 2017

  • Chow-Liu algorithm now supports networks with a mixture of discrete & continuous nodes (hybrid)
  • TAN algorithm now supports networks with a mixture of discrete & continuous nodes (hybrid)
  • Batch discretization of columns. This improves the performance when discretizing many columns
  • Fixed IsConstant bug in 'Add nodes from data wizard'. Only affected the user interface.
  • Added progress and cancellation to 'Add nodes from data wizard' (VariableGenerator in API).

Version - Monday, May 1, 2017

  • New Search & Score structural learning algorithm
  • Memory leak in inference engines fixed
  • Detect integral floats option added to variable generator (Add data from nodes)
  • New 'Add nodes from data' wizard added to User Interface.
  • New 'Message Bar' added to 'Network Viewer' in the User Interface that guides new users
  • Variable generator (Add nodes from data) detects constant variables

Version - Wednesday, April 5, 2017

  • Mutual information calculator - User Interface
  • TAN structural learning now supports continuous variables
  • Chow-Liu structural learning now supports continuous variables
  • Application shortcuts now uninstalled correctly when previous major versions are installed.

Version - Sunday, March 12, 2017

  • Tree Augmented Naive Bayes (TAN) structural learning algorithm - discrete nodes supported
  • Chow-Liu structural learning algorithm - discrete nodes supported
  • Entropy now supports conditional and joint entropy as well as marginal entropy - API support
  • Mutual information - marginal, joint and conditional - API support

Version - Friday, February 17, 2017

  • Help is now only available online
  • Parameter count: The parameter counter is no longer restricted to 32 bit integers (Table sizes still are).
  • Structural learning: numerical stability fix within statistical tests causing 'Failed to converge with values...' error
  • Express edition: fix for networks with noisy nodes exceeding node limit in express edition.
  • R squared combiner: error improved when weighted case counts are zero

Version - Sunday, November 13, 2016

  • Fix for FIPS enabled machines (Federal Information Processing Standard)

Version - Sunday, October 23, 2016

  • Save as image: allows saving a network to an image
  • Structural learning: fix for discrete 'variables' with a single state
  • Histogram density now throws sensible error if all cases are zero weighted
  • Structural learning: fix for extreme cases which result in negative degrees of freedom

Version - Saturday, July 23, 2016

  • Parameter tuning
  • Sensitivity to evidence (one way)
  • Sensitivity to evidence (two way)

Version - Monday, June 20, 2016

  • Online learning (Adaptation)
  • Histogram based empirical density, which can be used to model the distribution of log-likelihood values
  • Auto-insight now supports multiple variables and multiple states in the user interface and supports discrete nodes in hybrid networks.
  • Fixed re-enabling of buttons in the user interface after longer query times.
  • Added R squared to regression statistics
  • Decision graph algorithm selector added to user interface
  • Decision graph strategy (policies) can now easily be queried from the user interface
  • Set all values option added to distribution editor

Version - Tuesday, May 24, 2016

  • Fixed reading Bayes Server files saved from Java, and then loaded in .NET. Only affects networks with State values.
  • Auto insight target variable and state split in the User Interface to make selection faster and easier.
  • New Time Series learning mode.

Version - Saturday, April 30, 2016

  • WindowDataReader GetOrdinal/getColumnName bug fix.
  • 'Add nodes from data' now supports Decision and utility nodes.

Version - Saturday, April 23, 2016

  • Simple in-memory data store (DataTable, DataRow,...) added to Java API. Allows easier integration of data driven features from R (Data Frames), Python (Data Frames) & Matlab.
  • Data windowing added to APIs (WindowDataReader, WindowDataReaderCommand).

Version - Monday, April 11, 2016

  • Default log-likelihood tolerance changed to 1e-6
  • BREAKING CHANGE: Java IDataPartition interface renamed to DataPartition.
  • BREAKING CHANGE: Java DataPartition & EvidencePartition moved to namespace.

Version Alpha - Saturday, March 19, 2016

  • Fixed csv connection editors in the user interface

Version Alpha - Friday, March 11, 2016

  • Help added for Decision graphs (LIMIDS)
  • CSV support added to user interface.
  • New Oil Wilcatter sample decision graph added.
  • Walkthrough data now included in an Excel spreadsheet.

Version Alpha - Saturday, March 5, 2016

  • Decision graphs (limited memory influence diagrams LIMIDS)
  • Multivariate conditional utility queries (can be used to generate data for profit/cost surfaces).
  • Decision graph parameter learning.
  • Utility variables with variances.
  • New Expert disagreement sample network added.
  • Decision graphs with both discrete and continuous probability (chance) nodes.
  • BREAKING CHANGE FROM 6: The default convergence method for parameter learning is now log-likelihood.
  • BREAKING CHANGE FROM 6: The default prior for parameter learning is now uniform.
  • BREAKING CHANGE FROM 6: The default initialization method for parameter learning is now clustering.
  • BREAKING CHANGE FROM 6: ParameterLearningOptions.Tolerance now defaults to null, with the default depending on the Convergence method used (available via ToleranceOrDefault).

Version - Saturday, January 30, 2016

  • Detect cluster count now supported in the user interface.
  • Fix for clustered initialization with weighted data during parameter learning.

Version - Friday, December 25, 2015

  • Fix for disabled menu items in the User Interface, caused by a bug in .NET 4.6.1
  • Fixed NaN error which occurred during clustered initialization with a variable with zero variance.
  • Fixed 'Specified argument was out of the range of valid values' error in structural learning caused by zero degrees of freedom in test.

Version - Saturday, October 10, 2015

  • API: Detect the number of states of a discrete latent variable (cluster/mixture).
  • API: Convert from a covariance matrix to a correlation matrix.

Version - Sunday, October 4, 2015

  • Conditional queries such as P(A|B) and P(A,B|C,D) now directly supported in the API and User Interface.
  • Soft evidence can now be adjusted by existing distributions or queries.
  • List of errors now available in the Confusion matrix user interface.
  • Checkbox bindings in 'Add nodes from data' dialog fixed.
  • Network viewer theme mordernized.

Version - Sunday, August 9, 2015

  • Performance of multi-variate Gaussians improved.
  • API: New instantiation methods on CLGaussian, that provide improved performance
  • Structural learning now correctly supports discrete variables with only a single state

Version - Saturday, June 27, 2015

  • Avoid 'Covariance matrix is not positive definite' errors during structural learning, with numerical stability enhancements.

Version - Saturday, March 28, 2015

  • Add multiple states to a variable.
  • Further improvements to the numerical stability of high dimensional Gaussian distributions.
  • Gaussian distribution plots can be weighted, to decrease visibility of lower support Gaussians.
  • Temporal scatter plot - fix for variables plotted on incorrect axes.

Version - Wednesday, March 25, 2015

  • Query plots now support ordering of variables.
  • Clustered initialization - performance improvements with large discrete variables.
  • Clustered initialization - handling of temporal variables improved.
  • Clustered initialization - improved handling of conditional latent variables.

Version - Saturday, March 21, 2015

  • Clustering initialization method fixed to work with distributed parameter learning.
  • Numerical stability of high dimensional Gaussian distributions improved.

Version - Saturday, March 14, 2015

  • New parameter learning initialization algorithm [Clustering], that often provides better results with fewer EM iterations. By default this is not enabled, to maintain compatibility with version 6.
  • Rendering of Gaussians (visualizing queries and parameter learning monitoring) improved to handle 1000s of Gaussians.
  • Initialized state now output at iteration 0 during parameter learning progress notifications.
  • Random sampling parameter learning initialization fix that provides improved starting covariances.

Version - Tuesday, February 10, 2015

  • Performance improvements to Relevance Tree inference with large discrete variables.
  • Fixed incorrect max propagation log-likelihood calculation for certain Relevance Tree inference queries.
  • Performance improvements to parameter learning.
  • Fixed occasional incorrect display of the last variable in the Data Map window in the user interface.

Version - Sunday, October 5, 2014

  • Uniform prior option added to parameter learning user interface
  • Data tables are now cached by function in the user interface.
  • New variable argument Node constructor added for creation with variables.

Version - Thursday, September 25, 2014

  • Automatic check for updates now has limited support for automatic proxy detection.
  • Fixed Table.getMaxValue in Java API. Previously properties were hidden.
  • Fixed information columns filter in data map window
  • Fixed data explorer bug caused when non temporal data was not mapped when temporal data is.
  • Improved the performance of Node explorer for discrete variables with large numbers of states.

Version - Thursday, September 18, 2014

  • Improved inference performance with large discrete variables.
  • Improved parameter learning performance with large discrete variables.

Version 6.3.9141.0 - Wednesday, September 3, 2014

  • Documentation updates.

Version 6.2.9118.0 Beta - Thursday, August 21, 2014

  • Added selection likelihood per pane in auto insight viewer.
  • Documentation updates for new features

Version 6.2.9111.0 Beta - Tuesday, August 19, 2014

  • Fixed scrolling in auto insight viewer.

Version 6.2.9110.0 Beta - Sunday, August 17, 2014

  • States are now sorted in evidence explorer.
  • Performance of auto insight user interface improved.
  • Minor breaking changes to distributed parameter learning method arguments.

Version 6.1.9099.0 Beta - Monday, August 11, 2014

  • Uniform discrete priors
  • Fixed log-likelihood monitoring in distributed parameter learning
  • Confusion matrix grid view

Version 6.0.9068.0 Alpha - Thursday, August 7, 2014

  • Save and load Networks from strings.
  • Fixed MarginalizeLowMemory in MPE mode.
  • Node, variable and state lookup by name can optionally throw if not found.
  • Many new methods for easier access to Table and CLGaussian parameters, using Variables and States directly.
  • New methods to facilitate easier creation of Nodes, Variables and Tables using variable length parameter lists.

Version 6.0.8976.0 Alpha - Tuesday, July 29, 2014

  • Excel batch queries fix
  • Batch query create table data types fix

Version 6.0.8931.0 Alpha - Thursday, July 24, 2014

  • Auto insight decision tree

Version 6.0.8901.0 Alpha - Sunday, July 20, 2014

  • Fixed check for update code for pre-releases.

Version 6.0.8899.0 Alpha - Saturday, July 19, 2014

  • Distributed parameter learning
  • Auto insight
  • Breaking change to EvidenceReader interfaces.

Version 5.6.8979.0 - Tuesday, July 29, 2014

  • Fixed check for update code for pre-releases.
  • Fixed bug in Network.Load
  • Batch query create table data types fix

Version 5.5.8753.0 - Saturday, March 8, 2014

  • Fixed query explorer repaint issue.
  • Zoom slider improved.
  • Fixed create tables sql generation - state indexes option

Version 5.4.8708.0 Beta - Saturday, February 22, 2014

  • Fixed bug in the batch query user interface. Discretized variables were not available when constructing a confusion matrix.
  • Improved the numerical stability of the structural learning algorithm, especially with highly correlated data.
  • Many new help topics and new feature help completed.
  • New buttons to facilitate easier adding of default Case, Weight and Time column names.
  • User interface now defaults to times not being zero based indexes.
  • New noisy node sample network
  • New walkthrough - Discovering insight using comparison queries

Version 5.3.8627.0 Beta - Sunday, January 26, 2014

  • Fixed additional bug in 'Add nodes from data' which lead to an 'Object reference not set to an instance of an object' error.
  • Improved equal frequency discretization algorithm with handling of integral types.
  • Improved equal frequency discretization algorithm to more often return the suggested bin count with heavily skewed distributions.
  • Fixed misleading error reporting when invalid data is detected during concurrent parameter learning.

Version 5.2.8606.0 Beta - Monday, January 20, 2014

  • Cross validation added to API
  • Residual plots added to batch query window
  • Fixed bug in temporal structural learning
  • Fixed timeout bug when writing large batch query predictions to a data source
  • Fixed State Not Found Action bug
  • Fixed bug in 'Add nodes from data' which lead to an 'Object reference not set to an instance of an object' error.

Version 5.1.8329.0 Beta - Wednesday, October 2, 2013

  • Structural learning
  • Java API

Version 5.0.8118.0 Alpha - Sunday, September 1, 2013

  • Alpha expiry date extended

Version 5.0.7738.0 Alpha - Wednesday, May 29, 2013

  • Loopy belief approximate inference algorithm
  • Likelihood weighting approximate inference algorithm
  • Value of Information
  • Comparison queries
  • New data discovery options such as maximum state probability and sort ascending
  • Tree query (tree width)
  • Noisy nodes (Breaking changes with Node.Distributions)
  • Data Count
  • Data sampling now outputs weight.
  • Equal frequency discretization algorithm improved.
  • Case sampling
  • Entropy
  • Nested DataReader in API
  • New visuals for sequence data(e.g.Markov models, Hidden Markov models, Click modelling)
  • Query-order states
  • Manual re-order states

Version - Saturday, February 23, 2013

  • In a few cases, this fix solves early converegence issues in the parameter learning algorithm

Version - Friday, January 18, 2013

  • Query explorer refresh problem when .NET 4.5 is installed
  • Case limit for express edition changed to evaluate per iteration

Version - Saturday, September 22, 2012

  • The 'Time is zero based index' checkbox in the 'Data Map' window now works correctly
  • Variable.StateValueType now raises change notifications
  • Numerical stability improved in Bayesian networks with continuous variables, to avoid the exception 'The multiplication is undefined' being raised.

Version - Thursday, June 7, 2012

Version - Thursday, April 26, 2012

Version - Tuesday, March 20, 2012

Version - Wednesday, February 1, 2012

Version - Saturday, December 17, 2011

Version - Saturday, June 30, 2012

Version - Sunday, May 8, 2011

Version - Friday, February 18, 2011

Version - Saturday, January 8, 2011