Software Release Notes 4.6
by Markus Lauterbacher
For our spring release 2020, we have focused on completely new features and improvements in usability.
We have been working on a new concept for model evaluation in Brainer for a while now. After an appropriate testing phase, many new features are now officially available. Plots and evaluation functions have been cleaned up and a new, powerful GUI enables interactive, transparent evaluations. Read more in the Brainer User Guide at Postprocessing.
In Stipulator, we are able to present new features: a plot for displaying different case sets, a plot for visualizing channel (-meta) information, and a general feature for interactive coupling of plots.
Cross-tool improvement of GUI usability: This includes, among other things, a revision of modal dialogs, as well as indicators for ongoing operations.
Bug fixes: We encounter some problems when working with our tools ourselves. In addition, we are very grateful that our support channel is always supplied with important inputs regarding usability, bugs and desired features.
Another note for developers: On our roadmap a high-priority issue is, that we want to define several data sources in the Stipulator for one case. For this purpose, there will be interface adjustments in Stipulator in the next release.
In the following, we list the key new features and fixed issues introduced in this release in relation to the previously released version 4.5.
- (STP-607) A new plot for the visualization of set operations for groups, selectors and categories was implemented. It is accessible via the new menu Analysis > Cases > Count... . More information in the Stipulator User Guide at Data analysis > Case oriented analysis.
- (STP-549, STP-550) Another new function was introduced in the new menu Analysis > Cases. The menu point Channel info... provides channel meta information for the selected cases and their channels. This allows for a quick overview with respect to time of first and last sample, duration, sampling rate, number of samples, memory consumption, missing channels and outsourced-state. More information in the Stipulator User Guide at Data analysis > Case oriented analysis.
- (STP-656, STP-667) It is now possible to interactively link different analysis plots. The menu Tools > Link case highlighting to figure links an interactive highlighting to another figure's selection (via MATLAB data brush) and Stipulator selection respectively. Vice versa Stipulator may follow a figure's case selection (Cases > Select > By figure (linked selection)). It will select cases, which were highlighted with MATLAB's data brush in the linked plots. This feature is supported by the following plots: Scatter, Signal plot and CDF.
- (STP-600) The plot function Analysis > Plot attributes > Correlation matrix will remember its last configuration.
Performance, robustness and usability
- (SST-149, SST-159) Improved performance in the Signal Processing GUI with respect to network drives.
- (STP-491) A consistency check after loading a Stipulator project was added.
Adjustments to the command interface
- (SST-150) Command 'calcsignals' will now apply Stipulator's current procesing level, if there was no explicit processing level passed. Before it was processed.
- (STP-662) Setting case selections via CLI (e.g. via 'setselection') would order the passed indices, so that subsequent operations (e.g. 'calcsignals') were applied in different order. This bug was fixed. Selections will no longer get ordered when set.
- (STP-221) Command 'selectbyexistingfile' is now applied to shown cases instead of selected cases. Furthermore it returns case indices instead of selection indices.
- (STP-452) Command 'getattributenames' will return a row vector now instead of a column vector.
- (STP-521) Command 'enablelogging' was removed.
- (STP-622) The new command 'getactionnames' will return all action names of the Stipulator project.
- (STP-637) When sorting cases by column "Channels", the number of found channels is now taken as reference for the order.
- (STP-664) Refresh view does not ignore the current order of cases any more.
- (STP-638, SST-147) Plotting actions (Actions and signals vs. time) with Inf-values may have caused an issue with missing fire-patches in the figure. This bug was fixed. The location of action labels in the plot was decoupled from the actual action information.
- (STP-626) Operations in the side panel do not lead to new empty figures any more.
- (STP-665) Attribute extraction did so far not check for unique names. The 1-of-C encoding of categories could introduce duplicate names. This error source is now considered.
- (STP-648) A bug was fixed where the side panel's group selection would show/select cases of the same groups differently.
- (STP-635) The command 'removeattributes' crashed, if existing and non-existing attribute names were passed. This bug was fixed.
- (STP-643) A bug was fixed where the set option "Plot/export perturbations" led to non working other plot options (e.g. by category).
- (STP-644) A bug with plotting perturbations was fixed. On processing level raw there will no longer be depictions of perturbations.
- (STP-490) A bug was fixed where Stipulator's display configuration could reach an inconsistent state.
- (STP-627) A bug at version upgrade related to cache management was fixed.
- (STP-630) Changing link paths may have caused reordered link values. This bug was fixed.
- (STP-649) Missing functionality of command 'addsearchstringfromcase' was complemented.
- (MISC-156) A bug was fixed, where cases with NaNs in actions could not be plotted.
- (STP-56) A bug in pulse extraction with less than two channels was fixed.
- (STP-284) File > Export > Atrributes... now requires an installed MS Excel to allow for export to its format.
- (STP-250) A help button was added to the Edit case GUI.
- (STP-639) Action shift at Extraction > Properties... is no longer supported.
- (BRN-226, MISC-171) Completely refactored model evaluation
For usage details, see The Brainer User Guide section Postprocessing. Key features:
- Available data, models and evaluation plots are visible in a new "postprocessing" user interface.
- Each plot now has individual settings and an exemplary icon as a preview.
- Creation, management and exchange of so-called plot profiles, which also improve the reproducibility of evaluation results. For example, a profile can contain the same plot with varying settings, or any number of different plots with one setting configuration at a time.
- Depending on the plot, the evaluated models and/or outputs can be (de-)selected interactively.
- Performance: calculated model responses or valuation measures are now cached and reused in the next evaluation (e.g. creation of one or more plots) if data and model selection do not change.
- (BRN-320) Data import with already existing data was refactored. Especially the option merge was improved to offer greater usability and transparency. More information and examples in the Brainer User Guide.
- (BRN-318) The plot show missing values was complemented to account for all non-finite values in the DataMatrix.
- (MISC-201) Plots with a lot of data to visualize will warn the user beforehand about possible visualization performance issues.
- (BRN-351) A bug in the transformation GUI was fixed: When multiple parameters were selected, the resulting histogram of transformed data did not depict correct information. Furthermore copying of coefficients if multiple parameters are selected was disabled.
- (BRN-438) A bug was fixed where DMs with perturbations would not allow to set post transformations.
- (MISC-153) A bug in the confusion matrix was fixed. Having selected only same-class target datablocks may have led to wrong labels.
- (BRN-291) An error with the learning rate plot was fixed. Furthermore it is now available for LSTMs, too.
- (MISC-191) Training of LSTMs will now use validation data defined by Brainer's sampling.
- (BRN-447) A bug with Stratification was fixed.
- (BRN-457) A bug on removing data blocks at Data matrix > Edit... was fixed.
- (BRN-463) A bug was fixed, where exporting model responses failed.
- (BRN-304) A Bug with confusion matrix and multi class problems was fixed.
- (BRN-454) Selecting the same performance measure at "Select models by performance..." is no longer possible.
- (DM-50) Plotting a DataMatrix's raw data will ensure that there is exactly one variant (i.e perturbation) preselected.
- (MISC-184) Improved consistency of terms Time and Abscissa throughout plots and settings depending on the problem type.
- (UG-98) In the Brainer User Guide the section postprocessing has been renewed, restructured and extended:
Main section Postprocessing: Introduction to the scope and aspects of postprocessing as well as the explanation of the new user interface.
Subchapter Postprocessing plots: Explanation of the new interaction panel, overview of all available classification and regression evaluations, as well as description of the settings and contents of each plot.
Subchapter Postprocessing performances: Introduction to the evaluation concept, listing of available evaluation groups, as well as further descriptions of conventional measures (e.g. MSE) and classification performances.
Subchapter Postprocessing customization: Description of the object-oriented extension of custom evaluation plots and evaluation measures for developers.
- (BRN-320) Data import with already existing data was refactored. Especially the option merge was improved to offer greater usability and transparency. More information and examples in the Mapper User Guide.
- (EXP-344) A bug was fixed where importing multiple data with inhomogenuous structures lead to an error.
- (EXP-347) When filtering the expecation list, it was possible that settings of the wrong expectation were shown or edited.
- (ALL-230) All dialogs in windows were checked for their modal property and adjusted if necessary. The currently relevant input window will remain on top of the application.
- (ALL-231) The tools indicate their busy state while saving/loading projects.
- (ALL-190, ALL-192) Adjusted log configuration and output formats.
- (CON-13) A bug with pulldown-inputs was fixed. They were shown cropped under certain conditions.
- (SST-153) The function ssplot will return a legend handle as the second output argument.
- (SST-154) The function ssplot will no longer create a new figure automatically.
- (SST-134) A bug in ssmovingavg was fixed.
- (SST-151) The function ssfft will no longer fail on empty channels.
- (SST-135) New ss-filter function ssrotated2d.
- (SST-155) New ss-filter function ssimplode.
- (SST-157) Listing channels by name as cell array in the signal processing GUI will tolerate line breaks.
- (SST-90) Improved usability of sswrite2csv.
- (SST-103) Desupport of ssaddattr. Use sssetattr instead.