(Istar is the new name for what used to be called KBTools. KBTools was considered a rather boring name, so we have substituted one that is rather corny, and perhaps cheeky, if you are a Tolkien addict.)
A package for building knowledge based systems and then running them. It is of the inference net kind, with both backward and forward chaining. It is similar to the rule based kind but more flexible.
KBSs make inferences from input information. Usually running them involves asking the user a sequence of questions until enough information is gathered to come up with an answer. Microsoft Wizards are a simple version of this.
This version is a 'draft', letting you get the feel of it. Other things will be added in future to make it more usable.
To install, download istar.lha to a file on your hard disk or floppy, uncompress using lha and then double click on the Install_Istar icon.
Istar should run on any Amiga with at least 2MB of memory and Workbench 2.0 or later. It has been tested on the following:
Amiga 1200, 2MB, Workbench 3.0.
Amiga 4000, 6MB, Workbench 3.0.
There is possibly a slight bug on 4000's, see limitations and bugs. Istar is untested on Workbench 2.0.
There are two levels at which to learn to use Istar: 'driving' the features of the software and using it to do something be useful to you. These are explained in the files:
Read them when you are ready to start. In addition the following files give more detailed information:
There is still much documentation to be written, and not all the facilities of Istar are explained in the above files. But Istar should be reasonably robust, so you can experiment.
# Graphical input. Most KBS software require you to express your rules or inferences using text such as:
IF ?requested_class = PENICILLINS AND Patients_age >= 0.8*Allowable_age THEN Patient's age is acceptable for the new drug.
which is sensitive to syntax errors. But with Istar you express your inference graphically, as boxes and arrows. To build your KB you draw boxes and link them with arrows. Then, at your convenience, fill in the detail. (Some KBS software such as NExpert can display a box and arrows diagram that has been entered textually, but with Istar the main extry medium is drawing.)
# Smooth user interface. Most software that allows you to draw box and arrows diagrams is clumsy. For instance, to link two boxes:
Click on Box1. Mouse up to Toolbar or Menu Select 'Link boxes' Click on Box2. (Draws straight line between them that crosses others) Mouse up to Toolbar or Menu Select 'Bend link' Click on position of bend etc.
In Istar you start drawing from edge of Box1, hitting space wherever you want bends and releasing over Box2. No menues or toolbars in sight. The difference is even more marked when you want to redirect a link from one box to another.
# Easy exploration of KB. Find by Name. Also show a box with all its antecedents, all attached, etc. So you can find out what input information is will have an effect on a given item.
# For knowledge generation. This means the user interface actions do not interrupt your flow of thinking as you express your knowledge. Just as "an architect thinks with his pencil" so you can "think with the mouse".
# Range of types. Normally you are offered integer, float, string and perhaps dates. With Istar you get other types too: probabilities, bayesians, proportions, ratios, odds, directions, and several others. In Istar we attach the semantics to each type (e.g. in ratios and odds, the greatest common divisors are automatically cancelled out). Others to be added soon, such as IFF bitmap, anim, sound sample.
# Range of useful inference methods. Not just add, subtract, multiply, divide, AND, OR, but also things like Chhoser, First-Known, First-OK, Number of Answered, Concatentation, Is-In, etc.
# Kernel written in assembler, for speed and robustness.
Inference is working out the value of an attribute from the values of other attributes.
Backward chaining is the searching backwards (from right to left on most box and arrows diagrams of inference nets) to find the next relevant question to ask.
Forward chaining is the propagation (from left to right) of new information. Like recalculation in a spreadsheet.
The current version of Istar is only a 'draft'. It is being presented to let you get the feel of it. But in future you will have scripts, demons, ARexx, PostScript output, dumping of the KB to PROLOG, etc.
Copyright (c) Andrew Basden, 1996