The major improvement of version 4.33 in comparision to version 4.3 concerns the tool documentation. User Manuals for Stipulator, Brainer, Mapper and Expectator have been rewritten, significant parts of them completely from scratch, and thoroughly reviewed. Updated screenshots and examples have been included, as well as a more user friendly navigation. These updated User Guides are intented to make it easier to unleash the power within our tools.

Attention has been paid to tool installation issues as well: a more intelligent integration on existing startup.m files is now available and each tool will issue a warning in case multiple installations are found in the MATLAB path at the same time.

In the following we list the most significant new features introduced and issues fixed in this release.

Stipulator

• Added option to save calculated distance matrices to a file.
• Extended functionality of Links > Change path..., which now can replace any substring in the middle of a link and not only at the beginning.
• Plot preferences are applied to all Stipulator plots.
• Default value for t0 for a new case changed from 0 to -Inf.
• In a new Stipulator file, extraction operates only on selected cases by default (existing files take into account the setting "Only extract selected cases" in Extraction > Properties... as usual).
• Fixed issue with plotted actions not always matching set action times.
• Fixed issue when a user has no write permissions for a Stipulator LOG file.
• Fixed issue with directory depth to find files on Cases > Import...
• Fixed issue when deleting all reference channels.
• Fixed issue with interactive selection on attribute scatter plots.
• Fixed issue with add and delete cases from groups not saving changes.
• Fixed issue when concatenating two cases.
• Fixed issue when modifying actions with set mode.
• Improved performance of several distance calculation functions through parallelization.
• Stipulator extraction functions have been refactored to ensure full compatibility with the data matrix objects already introduced in release 4.3.

Brainer

• Added option to export tree based models to a MATLAB M-File.
• Improved usability of Data > Analyse > Plot inputs.
• Removed support for obsolete model type classregtree; current tree models are based on MATLAB functions fitctree and fitrtree, both available since MATLAB R2014a.
• Removed support for WEKA based model types (support vector machines, classification trees, ...), since all of them are directly supported by MATLAB Toolboxes now.
• Removed support of user defined tansig_approx for Backpropagation models.
• Fixed issues with plots "Model response in input space" and "Times of Classification relative CDF per group".
• Fixed issue in training with selected feature set after a Feature Selection process.
• Fixed issue when exporting model responses to TXT or XLS data formats.
• Fixed issue with models not showing after switching problem type.

Mapper

• Added Data > Analysis options as available in Brainer: "Plot inputs", "Plot raw data" and "Plot missing values".
• Improved performance on plotting parameter values.
• Removed obsolete training settings Goal, Maximum failures and Minimum gradient.
• Support resizing maps in external figures.
• Support using white spaces in parameter names.
• Fixed issue with updating parameter transformations after removing parameters.
• Fixed issue when using Postprocessing > Clustering.
• Fixed issue when adapting several maps at the same time.
• Fixed issues when plotting maps with several Mapper files open.

Expectator

• Menu Expectations > Groups > Define from selection... does not open edit groups dialog any more, it only asks for a group name for the new expectation group. This behavior is consistent with data groups.
• Data sources can be defined in a distributed text file called DataSources.txt: all occurrences of such a file in the MATLAB path are gathered by Expectator to generate the list of available data sources.
• Refactored all sort and reorder options into the View menu to avoid duplicated entries.

General Improvements

• Performance optimization of Matrixplot (using imagesc instead of patches)