Version 126.96.36.199 - Sunday, September 16, 2018
- Pattern analysis helps differentiate a discrete target state from the other states in the same variable. Useful for understanding clusters in a cluster/mixture model.
- Fix for removal of JAXB from Java SE 9/10 (Java breaking change)
- Data mapping editor updated
- BREAKING CHANGE: .NET 3.5 no longer supported
- BREAKING CHANGE: Java 6 no longer supported
- BREAKING CHANGE: Entropy class removed from Analysis namespace. Use more advanced version in Statistics namespace.
- BREAKING CHANGE: Cross validation classes moved to Data namespace
- BREAKING CHANGE: X, Y, Width & Height properties on node class, which were already deprecated, now removed. Use Bounds instead.
- BREAKING CHANGE: ValueOfInformationOutput MutualInformation and Reduction properties renamed to support both MI and MEU outputs.
- BREAKING CHANGE: Excel files pre 2007 no longer supported by default User Interface connectors.
- BREAKING CHANGE: DetectInteralFloats in VariableGenerator now defaults to true
- BREAKING CHANGE: DataPartitioning now includes PartitionCount
- BREAKING CHANGE: Parameter learning TimeSeries mode now needs to be explicitly set to either Pinned or Rolling with API. Previously defaulted to Pinned.
- BREAKING CHANGE: Parameter learning with Decision nodes now requires DecisionPostProcessing to be explicitly set.
- BREAKING CHANGE: AutoInsightOptions.KLDivergence changed from bool to enum to allow for future additions.
Version 188.8.131.52 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
- Retracted Analysis for anomaly detection diagnostics
- Fix for 'Add nodes from data' with decimal data type (e.g. currency).
Version 184.108.40.206 - Wednesday, September 20, 2017
- Fix for MySql .NET connector in User Interface.
- Fix in user interface in parameter learning wizard with noisy nodes
Version 220.127.116.11 - Thursday, July 20, 2017
- Performance fixes to Impact Analysis and Log-Likelihood analysis with larger numbers of variables.
Version 18.104.22.168 - 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 22.214.171.124 - 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 126.96.36.199 - 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 188.8.131.52 - 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 184.108.40.206 - 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 220.127.116.11 - Tuesday, May 23, 2017
Version 18.104.22.168 - 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 22.214.171.124 - 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 126.96.36.199 - 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 188.8.131.52 - 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 184.108.40.206 - 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 220.127.116.11 - 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 18.104.22.168 - 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 22.214.171.124 - Sunday, November 13, 2016
- Fix for FIPS enabled machines (Federal Information Processing Standard)
Version 126.96.36.199 - 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 188.8.131.52 - Saturday, July 23, 2016
- Parameter tuning
- Sensitivity to evidence (one way)
- Sensitivity to evidence (two way)
Version 184.108.40.206 - 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 220.127.116.11 - 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 18.104.22.168 - Saturday, April 30, 2016
- WindowDataReader GetOrdinal/getColumnName bug fix.
- 'Add nodes from data' now supports Decision and utility nodes.
Version 22.214.171.124 - 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 126.96.36.199 - 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).
Version 18.104.22.168 - Saturday, January 30, 2016
- Detect cluster count now supported in the user interface.
- Fix for clustered initialization with weighted data during parameter learning.
Version 22.214.171.124 - 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 126.96.36.199 - 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 188.8.131.52 - 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 184.108.40.206 - 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 220.127.116.11 - Saturday, June 27, 2015
- Avoid 'Covariance matrix is not positive definite' errors during structural learning, with numerical stability enhancements.
Version 18.104.22.168 - 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 22.214.171.124 - 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 126.96.36.199 - Saturday, March 21, 2015
- Clustering initialization method fixed to work with distributed parameter learning.
- Numerical stability of high dimensional Gaussian distributions improved.
Version 188.8.131.52 - 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 184.108.40.206 - 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 220.127.116.11 - 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 18.104.22.168 - 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 22.214.171.124 - 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
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
- 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 126.96.36.19976 - Saturday, February 23, 2013
- In a few cases, this fix solves early converegence issues in the parameter learning algorithm
Version 188.8.131.5285 - 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 184.108.40.20697 - 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 220.127.116.1191 - Thursday, June 7, 2012
Version 18.104.22.16849 - Thursday, April 26, 2012
Version 22.214.171.12416 - Tuesday, March 20, 2012
Version 126.96.36.19997 - Wednesday, February 1, 2012
Version 188.8.131.5228 - Saturday, December 17, 2011
Version 184.108.40.20693 - Saturday, June 30, 2012
Version 220.127.116.1183 - Sunday, May 8, 2011
Version 18.104.22.16837 - Friday, February 18, 2011
Version 22.214.171.12421 - Saturday, January 8, 2011