About NeuNet Pro
Table Of Contents

About NeuNet

License Agreement

Specifications

How to Purchase

Version History

Future Plans

NeuNet Pro Specification Sheet

NeuNet Pro is a 32 bit program which requires Windows 95, Windows 98 or Windows NT4(sp3). Users may select from either the SFAM classifier or the classic Backprop algorithms.

Data Requirements

  • Data must be in a table contained in an Access 95 or 97 MDB file, up to one gigabyte in size.
  • The provided Import function may be used to convert comma seperated text files into the required MDB format.
  • NeuNet Pro can link most common databases if you have their ODBC drivers installed.
  • NeuNet Pro only reads the data file. Your data is never modified in any way.
  • Data may contain 3 to 250 fields and any number of records.
  • One data field must be set as primary key index.
  • Fields may be yes/no, byte, integer, floating point, date or text type.
  • When used in SFAM classification mode, prediction field may contain up to 256 different classes.

NeuNet Pro Feature List

  • Ease of use has been the ultimate goal in designing NeuNet Pro.
  • User may select which fields are to be included as neural net inputs (1 to 250).
  • User may select which field is to be the predicted neural net output (1).
  • User may select which records are to be used for training (10 to 32000).
  • User may select which records are to be used for testing (1 to 32000).
  • Training records and testing records may be re-defined at any time.
  • Records containing missing values will automatically be detected and handled.
  • Data may be imported from comma seperated ASCII text files.
  • Algorithms are fully compiled and optimized for extremely fast operation.
  • Users may select from either the SFAM classifier or the classic Backprop algorithms.
  • Several projects may be combined into a Multiple Neural Net.
  • Backprop algorithm allows one output value to be predicted using up to 250 input values.
  • Backprop allows up to 128 neurodes in the hidden layer.
  • SFAM classifier allows up to 256 possible classes.
  • SFAM algorithm uses up to 250 input values in determining class prediction.
  • SFAM allows up to 32,000 neurodes to be created in the hidden layer.
  • Browse through data table while comparing actual versus predicted.
  • Interactively experiment with field values while observing the effect on the prediction.
  • Data mine anomalies by performing a descending sort on difference between actual and prediction.
  • Perform graphical data mining by clicking mouse on scatter graph and confusion matrix.
  • While browsing data, rows may be interactively checkmarked for export.
  • While browsing data, rows may be interactively sorted by several fields.
  • While browsing data, rows may be located by field search.
  • Data normalization is automatic and transparent to the user.
  • Data normalization automatically centered to the most sensitive part of the sigmoid curve.
  • Data report shows minimum, maximum, average and standard deviation for each field.
  • A snapshot of the "best to date" neural net is automatically maintained.
    This snapshot allows "revert to best" and "resume from previous session".
    This snapshot allows the completed neural net to be embedded into other programs using the optional NeuNet Programmer's Kit.
  • Additional project snapshots may be saved by the user.
  • A comprehensive report (with graph) shows the statistical accuracy of the predictions.
  • Performance reports include scatter graph, time series graph and confusion matrix.
  • User may configure up to 18 recent projects to appear on the recent files list.
  • Context sensitive Help File responds to F1 key press in the main program.
  • Entire Help File may be easily printed, or printable manual may be downloaded from the web site.

A Complete Neural Network Development System

CorMac Technologies Inc.
34 North Cumberland Street ~ Thunder Bay ON P7A 4L3 ~ Canada
E m a i l