KnowledgeMiner
Simple, powerful statistical modeling shareware for your Mac
 
 
|Home| |Knowledge Discovery| |Publications| |Paper 1| |Paper 2| |Software| |Examples| |Registration| |Secure Online Purchase|
 
 
Download KnowledgeMiner 2.1 Here Free and Try It Out! 10/24/97
Download KnowledgeMiner
 

KnowledgeMiner Changes and Updates

 
Changes in version 2.2.1 - December/15/1997
 
  • 32k ASCII text import limit removed
  • fixed bug that causes a crash when calculating a large number of formulas in spreadsheet
  • fixed bug on PowerMac that does a wrong cell assignment when verifying a spreadsheet input by mouse click
  • improved and updated the text files and apple guide

Changes in version 2.1

  • tutorial updated.
  • removed bug that allowed opening two windows at once on ppc.

Changes in version 2.0

  • optimized for PPC and 68kFPU based Macs
  • PPC native app runs 8-10 times faster than 68k based version.Now it is possible to mine very large data sets for relevant relations or patterns within minutes instead of days.
  • Analog Complexing
    • available for the first time on a personal computer (Mac) as shareware.
    • shows automatically a prediction interval graphically (the calculated possible upper and lower limits) along with the most likely prediction of points. Now a prediction consists not only of sharp points but reflects also the inherent fuzziness and uncertainty of the objects. Applied to financial time series forecasting, e.g., this view focuses on the demand to consider the volatility of assets to get more reliable forecasts and decisions.
  • optimized modeling algorithms to make self-organization of structures and knowledge extraction from data far more effective than neural networks or statistics
  • Lite version
    • now has a larger table to work with data
    • two levels to setup modeling
      • a standard and
      • an advanced modeling setup dialog

KnowledgeMiner - Modeling Technology for the Rest of Us

KnowledgeMiner is a powerful, easy-to-use modeling and prediction tool which was designed to support the knowledge extraction process from data on a highly automated level.It works on two advanced self-organizing modeling technologies: Group Method of Data Handling (GMDH) and Analog Complexing. Built on the cybernetic principles of self-organization, KnowledgeMiner brings high-end modeling capabilities on your desktop without the need of being an expert in modeling since it will learn completely unknown relationships between outputs and inputs of any given system in an evolutionary way from a very simple organization to an optimal complex one by itself. The main advantages of this inductive approach are:
 
  • Only minimal, uncertain a priori information about the system is required. That means, even if you are not an expert in modeling, data analysis or designing a neural network you will be able to model, analyse and predict very complex objects of nearly any kind of system.
  • Very fast and effective learning process for ordinary PC's. That means, you can solve problems on your desktop in a reasonable time which you may have never thougth about before.
  • Modeling on very short and noisy data samples. That means, you can deal with a problem as is and don't have to construct artificial conditions for your modeling method to get it work.
  • Output of an optimal complex and cross-validated model. That means, you commonly can expect to get a model which is robust, as simple as possible and not overfitted. Overfitted models are not able to predict variables due to their reflection of random relationships between variables.
  • Output of an analytical model as a explanation component. That means, you can evaluate the analytical model to interpret the obtained results immediately after modeling. You don't have to guess why results are as they are.

In KnowledgeMiner you deal with data like in spreadsheets and, through its built in model base, you are able to create and store time series models, input-output models and predictable systems of equations (networks of input-output models) for each variable in one document. It is possible, for example, to create a linear or nonlinear time series model, a multi-input/single-output model and a system of equations for 20 output variables in one document. These models are easy applicable to sets of new data (prediction, classification, diagnosis) within KnowledgeMiner immediately. There is no need to import them as C code into other applications or to do other efforts to get them run.

KnowledgeMiner does not only do some useful work for you autonomously but also lets you the freedom to get some other work done in the same time by sending the complete knowledge extraction process into the background of your computer. For example, while I'm just writing this text KnowledgeMiner works very hard in the background of my computer to create an excellent prediction model of the US economy for me, hopefully (you can review the results if you have downloaded the package). KnowledgeMiner not only assist me to sieve some important information out of data but also helps me extremly saving my resources twice: by using the fastest and most robust knowledge extraction technologies available today and by giving me the chance to do some other work while it does its job.

 
The power and the advantages of KnowledgeMiner in comparision with statistical modeling tools and neural networks make it stand out as easier, faster and more applicable to a wide range of real-world problems. This makes KnowledgeMiner the least expensive and most effective modeling and prediction tool available on any platform.

 

KnowledgeMiner Lite Capabilities

  • spreasheet like handling of data including simple formulas and cell references
  • several built-in mathematical functions for extending the data basis:
    • xy, x(y/z), trigonometric, exponential and logarithmic functions, mean, sum and standard deviation, correlation analysis, random values, add uniform noise
  • opens ASCII text files
  • creates automatically
    • linear or nonlinear static GMDH-models
      • multi-input/single-output models as well as multi-input/multi-output models (system of equations) available analytically and graphically
    • linear or nonlinear dynamic GMDH-models
      • time series models, multi-input/single-output models as well as multi-input/multi-output models (predictable system of equations) available analytically and graphically
    • for up to
      • 50 input variables
  • enables background modeling
  • stores all created models in a model base dynamically
  • all models can be used for status-quo or what-if predictions, classification or diagnosis problems within KnowledgeMiner

KnowledgeMiner Pro Capabilities

KnowledgeMiner Pro is the modeling and prediction tool you need if you want to deal with complex, real-world problems contineously. It provides in addition to KnowledgeMiner Lite this features:
 
  • creates GMDH-models automatically for up to
    • 500 input variables (for more variables, call) enabling solution of complex real-world problems
  • it is possible to use a third data set, the examination set, additionally to training and checking sets for model performance validation during modeling. The validation results have influence on the selection and the network synthesis process.
  • creates nonparametric prediction models for fuzzy objects by Analog Complexing, an advanced pattern search technology for evolutionary processes. A synthesis of different prediction models (GMDH-based and Analog Complexing-based) is now possible as a powerful way to increase prediction accuracy.
  • features necessary to deal with large modeling problems more convinient

Unique Features of KnowledgeMiner

  • GMDH-type Neural Networks that perform
    • Active Neurons selecting their input variables themselves
    • advanced network synthesis and model validation techniques to end up in a robust, optimal complex model
  • creation of a best and autonomous system of equations (network of GMDH-type Neural Networks) which is ready for status-quo predictions of the complete system by default and which is available analytically and graphically (system graph) for results interpretation
  • a model base to store all models and to keep connected information together
  • Analog Complexing as a powerful pattern search technology to create predictions for fuzzy processes (the most market processes e.g.) which other methods may be not appropriate for.
  • completely autonomous modeling process which can work as background process on computers saving your resources either by working simultaneously with the modeling process or, for larger problems, by running the process overnight and getting some work done while sleeping

 

System Requirements

Software

MacOS 7.0 or later, Apple Guide, and QuickTime

Hardware

68k based Macs: 8+MB RAM, 68020 or higher with FPU (for a version that requires no FPU, please call)

PPC based Macs: 16+MB RAM, PPC 601 or higher

 

Contact:
julian@sierra.net
frank_lemke@magicvillage.de