May 16, 2019
pSeven 6.15 Release
DATADVANCE development team announces the availability of pSeven 6.15, a new version of our flagship platform for data analysis and optimization. This major update brings notable enhancements in design exploration and predictive modeling, and further extends pSeven’s optimization and integration capabilities. The release adds the support for categorical and discrete variables in various techniques and predictive modeling tools in Analyze, refines the existing linked selection functionality and contains various convenience features to take customer experience with the platform a step further. Read on for more information on these release highlights and for other details.
Design Exploration Enhancements
pSeven 6.15 enriches design exploration capabilities of the platform by introducing the redesigned Uncertainty quantification block and adding various improvements to the Adaptive design technique. The topical issues here include:
- Reworked and updated the UQ block, renamed to Uncertainty quantification. Some of its noteworthy features are as follows:
- Brand new configuration dialog similar to the Design space exploration block.
- SmartSelection support: the block automatically selects the most suitable analysis method based on your configuration of variables and responses.
- Support for advanced configuration scenarios including vector variables and responses, setting their properties dynamically using special input ports, batch evaluation mode allowing parallelization, and other.
- Results output as HTML-formatted reports that you can load into a Page viewer in Analyze.
- We added notable improvements to the Adaptive design technique which cover the following specific use cases:
- Significantly increased performance when generating a uniform high-dimensional DoE with no objectives but with a high number of linear constraints, which bound the DoE generation to a narrow subset of the design space.
- Better support for sample-based generation when response evaluation is disabled. The technique will now generate uniformly distributed points, avoiding clusterization.
- The Adaptive design for GP technique also received important updates:
- Improved technique performance and stability when generating a DoE with functional constraints.
- Added the support for multi-dimensional responses.
- Added parallelization support, so with a response model which can evaluate multiple points in parallel, the technique can work even faster.
Design Optimization Improvements
This update adds the following useful features to the Design space exploration block:
- Calculate and output a feasibility measure showing the degree of constraint violation for each design point.
- Enable additional ports which output the index of the current iteration. The index number can be used, for example, by a blackbox block to generate names of files that it creates.
These convenience features come on top of several updates in the internal optiomization algorithms, which noticeably increase the block’s performance in constrained optimization problems.
Predictive Modeling Enhancements
In this version, we continue to refine the functionality of the Model builder tool in Analyze. The highlights are:
- Functional and performance improvements in the Smart Selection mode, namely:
- Introduced the new, even more accurate and automatic method which decides whether to apply log transformation to the outputs data in the training sample. The new method essentially compares various models trained with and without the transformation and selects the best one. Log transformation may improve model accuracy when training outputs are exponentially distributed.
- Increased the SmartSelection algorithm quality and performance thanks to the latest updates in approximation techniques based on Gaussian processes (also supported by the ApproxBuilder block).
- New special model training mode, in which pSeven tests the training sample to find linear dependencies between responses, and trains a model which keeps these dependencies. This training mode is supported by the Model builder tool in the Smart Selection mode and the ApproxBuilder block.
- Load balancing and computational resource utilization improvement when training approximation models in parallel mode, which is default in SmartSelection.
Integration Block Updates
pSeven 6.15 extends the functionality of a few direct integration blocks.
This release continues to develop the Functional Mock-up Interface (FMI) support: we have added the capability to integrate Functional Mock-up Units (FMU) for Model Exchange into workflows using the FMI model block, and to export a pSeven model as a FMU using the ApproxBuilder block. Also, the FMI model block is promoted to release blocks and is no longer tagged as beta.
The STAR-CCM+ block, first introduced in pSeven 6.14, was updated with the new method to offload simulations to a remote host via SSH. In previous versions, the block could only connect to an already running remote STAR-CCM+ server, while the new method configures and starts the remote server automatically. Other improvements worth mentioning are:
- Added the ability to set global parameters directly from the block configuration.
- Added the ability to export simulation plots as images.
The last but not the least, the Excel block was updated to support paths to the Excel document and export files as inputs sent to the new Document file and Export file ports.
This means that you can dynamically load different documents to the block during a workflow run, or save each version of the working document as a new file, or select the new ports as workflow parameters so that the working documents can be quickly changed in Run.
Extended Categorical and Discrete Variables Support
This update extends the support for discrete and categorical variables:
- The Surrogate-based optimization and Adaptive design techniques now fully support discrete variables with non-integer values.
- Added the support for categorical variables in the SPLT and GBRT approximation techniques in Model builder.
- Improved handling models with categorical variables in the predictive modeling tools in Analyze.
More Synchronized Plots
pSeven 6.14 added the linked selection functionality to synchronize selections between Sample viewer, 2D plot and Parallel coordinates plot. pSeven 6.15 also adds Page viewer to this set of tools which support linked selection.
A Page viewer in the linked selection mode can show additional details for rows you select in a linked Sample viewer, 2D plot, or Parallel coordinates plot. You can also use it to thumb through the plotted design points, or to jump to a specific point by its index.
Custom Block Icons
As a convenience feature for users who design complex integration workflows, we have enabled custom icons for the Composite, Program, PythonScript, and Text blocks. These blocks often serve as wrappers for third-party tools, so making them stand out in a workflow often adds to its readability, helping in workflow editing and navigation.
This version has a new license format and therefore requires an updated license file. Licenses issued for the previous versions of pSeven are not compatible with pSeven 6.15 and above. If you are under active maintenance, you will receive your new license file shortly. In the case you did not receive an update, please contact firstname.lastname@example.org.
For the license update guide, see the License Setup section in the pSeven manual.
In this version, we have also resolved a number of issues and bugs. For a detailed list of updates, please see the release changelog. You can also contact us to get more information and pSeven updates.