.NET API documentation
Java API documentation
.NET API (full documentation)
Java API (full documentation)
Please note that the release history presented here only goes back as far as version 3.
- Tuesday, May 23, 2017
- 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.
- 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.
- 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).
- 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
- 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.
- 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
- 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
- Sunday, November 13, 2016
Fix for FIPS enabled machines (Federal Information Processing Standard)
- 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
- Saturday, July 23, 2016
Sensitivity to evidence (one way)
Sensitivity to evidence (two way)
- 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
- 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.
- Saturday, April 30, 2016
WindowDataReader GetOrdinal/getColumnName bug fix.
'Add nodes from data' now supports Decision and utility nodes.
- 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).
- 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 com.bayesserver.data.distributed namespace.
Version 188.8.131.52 Alpha
- Saturday, March 19, 2016
Fixed csv connection editors in the user interface
Version 184.108.40.206 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 220.127.116.11 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).
- Saturday, January 30, 2016
Detect cluster count now supported in the user interface.
Fix for clustered initialization with weighted data during parameter learning.
- 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.
- 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.
- 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.
- 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
- Saturday, June 27, 2015
Avoid 'Covariance matrix is not positive definite' errors during structural learning, with numerical stability enhancements.
- 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.
- 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.
- Saturday, March 21, 2015
Clustering initialization method fixed to work with distributed parameter learning.
Numerical stability of high dimensional Gaussian distributions improved.
- 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.
- 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.
- 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.
- 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.
- Thursday, September 18, 2014
Improved inference performance with large discrete variables.
Improved parameter learning performance with large discrete variables.
- Wednesday, September 3, 2014
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
Breaking change to EvidenceReader interfaces.
- Tuesday, July 29, 2014
Fixed check for update code for pre-releases.
Fixed bug in Network.Load
Batch query create table data types fix
- 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
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
New data discovery options such as maximum state probability and sort ascending
Tree query (tree width)
Noisy nodes (Breaking changes with Node.Distributions)
Data sampling now outputs weight.
Equal frequency discretization algorithm improved.
Nested DataReader in API
New visuals for sequence data(e.g.Markov models, Hidden Markov models, Click modelling)
Manual re-order states
- Saturday, February 23, 2013
In a few cases, this fix solves early converegence issues in the parameter learning algorithm
- Friday, January 18, 2013
Query explorer refresh problem when .NET 4.5 is installed
Case limit for express edition changed to evaluate per iteration
- 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.
- Thursday, June 7, 2012
- Thursday, April 26, 2012
- Tuesday, March 20, 2012
- Wednesday, February 1, 2012
- Saturday, December 17, 2011
- Saturday, June 30, 2012
- Sunday, May 8, 2011
- Friday, February 18, 2011
- Saturday, January 8, 2011