Improvements and Fixes in Recent Build

Enhancements

(NeuroSolutions)

Support added for Windows Vista
Code Generation and DLL support added for VC++ 2005

(NeuroSolutions for Excel)

Support added for Windows Vista

(Custom Solution Wizard)

Support added for Windows Vista
Added support for Visual Studio 2005

(NeuroSolutions)

HebbianSynapse did not learn until the number of weights were changed
Breadboards created with older versions of NS would occasionally crash due to the BackSynapse weights not getting properly allocated
Recurrent breadboards would sometimes crashed on load
SVM demo fixed to show a higher resolution image
Eliminated occasional crashes with demos
NeuralBuilder would lose selected input tags when the Randomize button was clicked
OLE Demo did not work properly

(Custom Solution Wizard)

VB Project Shells update: The VB Project Shells use DAO to read in the data files. DAO can only be used to read in data files with 255 or less columns. Some error checking was added to capture the case when the user's data file is too big and let the user know. The file "VBShell.vb" was updated.

VC++ Project Shells update: The buffer size in the function "FillArrayFromFile" of VC++ Project shells (in the file VCPPShell.cpp) was not big enough to read in large data files (large number of inputs). The buffer size was increased from 255 to 20,000.

Enhancements

(NeuroSolutions)

When the number of network weights change between runs, the Saved Error of the Criterion is reset to a high value so that the best weights will automatically get re-saved (if the Save Best switch is set).

(NeuroSolutions for Excel)

Test function now saves the original state of parameters it changes in the NeuroSolutions BB and restores then when it is done.
Added support for Excel 2007 (beta 2)

Bug Fixes

(NeuroSolutions)

Fix: Crashed when trying to copy and paste a subclass of BackArbitrarySynapse
Fix: Warning message regarding backsynapse weights not matching when trying to test a breadboard with LM learning.
Fix: Resetting network would occasionally crash the program.

(NeuroSolutions for Excel)

The following problems were fixed for single output 2 class classification problems (when running Test or LeaveNOut):

Fix: The confusion matrix was not being created properly when the first desired row was the larger numeric value of the 2 classes (i.e. if the first desired value was a 1 when the classes were 0 and 1).
Fix: The confusion matrix data was not being put in the correct place when the 2 classes were other than 0 and 1.
Fix: The Performance measures were not getting separated by class and they were not being calculated correctly.

Enhancements

Addition of a Welcome menu and cosmetic enhancements to the NeuroSolutions demos.

Bug Fixes

(NeuroSolutions)

Fix: Single step exemplar did not function properly when cross validation was active
Fix: When generating source code for unsupervised nets with the learning turned off at the control, the network would still learn.
Fix: Hybrid networks configured with LM learning did not update the weights until after the network was paused
Fix: Source Code License: NSLearn.cpp failed to compile under g++ (GCC) 3.3.5 due to multiple uses of for variables outside the for statement context. This code was changed to be compliant with the ISO standard.

(NeuroSolutions for Excel)

Fix: After training the first time, the user could go to the desired file and set it to normalize across whatever datasets they choose. The problem was that when there was a Production set in NS for Excel the user’s change of the normalization setting was getting lost.

Bug Fixes

(NeuroSolutions)

Fix: Number of processing elements of FuzzyAxon of CANFIS network changed to match version 4.0
Fix: Program crashed when the structure of input data files were changed between runs
Fix: False error message that reported the number of exemplars did not match between the input and desired files
Fix: The upper and lower bounds for the genetic parameters were not being saved with the breadboard.

(NeuroSolutions for Excel)

Fix: The table in the Testing report produced 2 columns for single-output function approximation problems.
Fix: Leave-N-Out did not save the best weights after the final training run.

(Custom Solution Wizard)

Fix: Version compatibility error reported with Visual Basic projects.
Fix: Error reported when trying to create a breadboard with a GeneticControl component.
Fix: Weights file saved by CSW could not be loaded back into the original breadboard.

Usability Enhancements and Bug Fixes

(NeuroSolutions)

Updated help text and sample breadboards for Code Generation examples
Fix: TestingWizard did not correctly set the "Display Every" parameter of the probes for breadboards configured for Iterative Prediction
Fix: A couple of Interactive Book examples did not run correctly in v5.0 of NeuroSolutions.

(NeuroSolutions for Excel)

Fix: "Overflow" error during when run "Test" under "Test Network".
Fix: "Translate Symbolic Columns" did not work when user logged in on a machine w/o Administrator privileges.
Fix: The "Percentage Correct" value was reported incorrectly for the Testing Report when a single 2-class column was used as the desired output.
Fix: Did not run properly under Excel 97.

(Custom Solution Wizard)

Support for NeuroSolutions for MATLAB v2.0
Fix: Minor differences in results between some NeuroSolutions breadboards and generated DLLs.

Usability Enhancements and Bug Fixes

(NeuroSolutions)

Addition of Levenberg-Marquardt learning algorithm
Addition of Double Precision Calculations
Addition of an "Annotation" column tag that will allow a row label to be included in the DataWriter output.
Improved SVM performance with the addition of a slack constant to the GaussianAxon.
Genetic Optimization log now written to a text file in addition to the display window.
Fix: Randomize of two separate files now produces more truly random files.
Fix: No longer crashes when user renames components on the breadboard before running the testing wizard.

(Data Manager)

Addition to Importing data from Access, Excel or text files.
Addition of various preprocessing and data analysis operations.
Addition to Load data directly into a NeuroSolutions Breadboard.
Addition to use the data to create a new neural network.

(NeuralBuilder)

Addition of a Randomize Rows button to create a randomized version of your data set.

(NeuroSolutions for Excel)

Leave N Out - Trains the network multiple times, each time omitting a different subset of the data and using that subset for testing. The outputs from each tested subset are combined into one testing report and the model is trained one additional time using all of the data.
Encode Two Class Column - The selected column of data is checked to verify that there are two classes contained within the column and is then encoded into another column. The data to be encoded can be textual or numeric, but must be translated to only numeric, integer codes. The encoded column will be written in the first empty column in the dataset.
Classification Report when testing on a single 2-class column - A classification report generated if the desired output of the testing set is a single two-class column.
Shift - The input data is adjusted to either move the inputs back by a specified shift value to do predictions or move the inputs forward to lead your desired output.
Pause Training - As an alternative to completely stopping a training run, the network can be paused, so training can be resumed at the same point.

(Custom Solution Wizard)

Addition of Double Precision Calculations support

(Usability Improvements)

Support for co-existence of NeuroSolutions v4.0 with NeuroSolutions v5.0
Remote Administrator support for hardware key

NeuroSolutions Home  |  Products  |  Downloads  |  Resources  |  Support  |  Order

Product questions? Contact info@nd.com
Website questions? Contact webmaster@nd.com
Interested in selling our software? Become a reseller!
Web Site Design and Implementation Copyright © 2008 NeuroDimension, Inc.

NeuroSolutions overview in Chinese.