DataEngine

Technical Details

Hard- and Software Requirements

  • PC with Pentium 300 processor
  • Microsoft Windows 95/98 or Microsoft Windows NT 4.0 or higher
  • 64 MB RAM
  • high resolution graphics board (800x600 or higher)
  • 12 - 47 MB free hard disk space
  • 32bit multithreading application
  • complete integration in Microsoft Windows operation system
  • An OLE interface allows to configure the models and to perform the training/labeling/test from within other programs, e.g. for Windows-based data analysis applications. This way DataEngine's functionalities can be called by any application program, regardless of the programming language.

Additional Software Requirements for Server License

  • Microsoft Windows 2000 Terminal Server

Top

Fuzzy Rule Base

  • linguistic If-Then Rules
  • examples of potential applications: knowledge-based diagnosis, classification, control and modeling
  • multi-step inference technique
  • supports representation of symbolic and/or linguistic information
  • fuzzy operators: minimum, maximum, algebraic product, algebraic sum, gamma operator
  • defuzzyfying strategies: mean of maxima, center of gravity and fuzzy output
  • debugging

Top

Multilayer Perceptron

  • supervised learning neural network
  • examples of potential applications: classification, modeling and control
  • learning rules: backpropagation, quickpropagation, Super SAB, resilent propagation, each with momentum and decay
  • learning rate decay to avoid overfitting
  • pruning algorithm for adapting the structure of the network architecture
  • shortcut connections are possible: direct connections of input with output parameters, skipping hidden layers
  • the 'best' neural network that appears during the training process can be saved automatically, the criteria for the 'best' network can be configured

Top

Kohonen Feature Map

  • unsupervised learning neural network
  • examples of potential applications: clustering and classification
  • labeling by examples
  • graphical representation of the feature map

Top

Fuzzy C-Means

  • fuzzy clustering algorithm
  • examples of potential applications: clustering and classification
  • initialize algorithm with pre-defined data partitions
  • labeling by examples
  • the 'best' number of classes can be determined automatically, the criteria for the 'best' number of classes can be configured

Top

Fuzzy Kohonen Network

  • unsupervised learning neural network
  • examples of potential applications: clustering, classification
  • very fast fuzzy training algorithm
  • labeling by examples
  • graphical representation of the feature map

Top

Data Editor:

  • spreadsheet-like usage
  • data table size only restricted by available memory

Mathematics

  • general (x, 1/x, sign, change sign, square, square root, power, +, -, *, Ö, factor, quotient, offset, cumulative sum, delta)
  • logarithms (exp, log, In)
  • trigonometric functions (sin, cos, tan, arcsin, arccos, arctan)
  • hyperbolic functions (sinh, cosh, tanh, asinh, acosh, atanh)

Scaling

  • normalization to 0..1
  • standardization
  • scaling to a specified range
  • linear transformation
  • denormalization

Missing Values

  • replace by constant
  • replace by minimum, maximum or mean
  • replace by previous or next value
  • linear interpolation
  • remove from data table

Import and Export of Data

  • import and export of Microsoft Excel files
  • flexible import and export of ASCII files
  • database interface including import and export of data via OLE DB/ODBC, 'mapping' allows to transform the values of categorical variables into numerical values
  • generic data acquisition board support (analog input and output, trigger and pre-trigger support)

Statistical Functions

  • general statistics (min, max, mean, variance, standard deviation, range, skewness, kurtosis, sum, sum of squares, number of values, number of missing values)
  • moving average, minimum, maximum, standard deviation and variance
  • histograms
  • correlation matrix
  • linear regression
  • the statistical batch analysis automates the first analysis of data by presenting high correlations and unpropitions distribution of values. This supports the identification of potential outliers and features with a high quota of missing values.

Signal Processing:

Fast Fourier Transformation

  • FFT, power spectrum, amplitude spectrum, phase and amplitude
  • several window functions (rectangle, triangle, Hamming, Hanning, Parzen, Welch and Blackman)
  • inverse FFT

Digital Filters

  • IIR-filters (Bessel, Butterworth, Cauer and Chebychev)
  • filter bands: high pass, low pass, band pass, band stop

Smoothing

  • simplified Least Squares smoothing (Savitzky-Golay)
  • degree of smoothing: 5, 7, 9, 11, or 13 points

Top

Project Window

  • serves as a framework for all files, models and feature information belonging to the same data analysis project
  • the alias concept allows to manage project files independently from the absolute directory path

Top

Graphing and Presentation of Results

  • standard text reporting facilities
  • presentation quality 2D and 3D charts
  • time series
  • scatter plots (2D and 3D)
  • line, bar, pie, step and combination plots (2D and 3D)
  • Hi-Lo Close plot
  • Gantt chart (2D and 3D)
  • bubble, contour, polar and radar plots (2D)
  • surface plot (3D)
  • various grid and scale options
  • easy access to all chart parameters

Top

Graphical Macro Language

  • easy-to-use graphical programming language
  • automation of pre-processing, data analysis and visualization
  • includes all the functions of the above sections Data Editor, Import and Export of Data, Statistics, Signal Processing
  • extendible by incorporating PlugIns (see below)

Top

Analysis component

Adjustment of operating

  • Feature values of objects can be varied and the model's changing outputs are displayed. A "slider" can be used to vary the values

Transfer function

  • On the basis of the chosen operating point a transfer function can be computed by varying (automatically) one or two features within their value ranges. The resulting function is visualized in a 2- or 3-dimensional diagram

Sensitivity analysis

  • On the basis of the chosen operating point a sensitivity analysis is carried out for the current model. To assess the sensitivity three measures are available. Each of them is computed by varying one of the features.
    These measures are:
    • the minimum and maximum values an output variable can have (shown as a hi-lo-close-plot)
    • the graph of an output variable (shown as a set of curves)
      the cumulative derivation of an output variable (shown as a bar diagram)

Test error

  • In addition to the test evaluations that exist in the model editors several additional error measures can be calculated here.

Matrix of confusion

  • For classification tasks a so-called matrix of confusion is computed to determine what kinds of misclassifications exist in the test data.

Top

PlugIn Interface

  • DataEngine is extendible by user defined functions based on MS Windows DLLs (Dynamic Link Libraries)
  • easy integration of other functions, techniques and programs into DataEngine
  • easy-to-use API for immediate production of results
  • several third party PlugIns available

Functions a DataEngine PlugIn should include:

void

FBLOCK_Create

(UserFB);

int

FBLOCK_Load

(UserFB, char* buffer);

unsigned

FBLOCK_Save

(UserFB, char* buffer);

void

FBLOCK_Copy

(UserFB, UserFB);

void

FBLOCK_Destroy

(UserFB);

int

FBLOCK_Configure

(UserFB);

int

FBLOCK_Check

(UserFB);

void

FBLOCK_Init

(UserFB);

int

FBLOCK_Execute

(UserFB, DEMatrix* data);

void

FBLOCK_Done

(UserFB);

Top

Online Documentation

  • outline of DataEngine
  • user manual
  • tutorials (including sample files)
  • theory (data analysis, fuzzy logic, neural networks)
  • programmer’s guide to user-defined function blocks

Top|Home|Contact Us|Info|Back