pyCogMod is a lightweight cognitive model testing API, based on the JESS rule engine. It's goal is to help researchers working with cognitive models, that need a fast, simple and reliable way to check a JESS-compatible production rule cognitive model on a given task, without having to worry about too much coding or low-level JESS functions. To use CogModAPI all the necessary JESS components of the cognitive model must already be defined, i.e. the working memory templates & configuration, the cognitive model itself. As an example, a simple cognitive model for solving simple algebra equations, together with all the necessary domain dependent functions and the JESS related structures, is provided for testing purposes. However, the provided API does not contain any domain dependent code and can be used for any domain (not just algebra).
Note: Cognitive modeling is a complicated task, and generating a cognitive model requires a series steps, including:
- coding to implement domain dependent functions (e.g. for algebra this may be divide two numbers, figure out if an equation has a coefficient etc),
- coding the cognitive model itself, which is a set of production rules that uses the domain depedent functions and describe how the task is solved (e.g. what to do when an equation is given)
- preparing whatever is necessary to setup the working memory of the rule engine that will "run" the cognitive model (e.g. an interface to perform the task, templates, initalization etc).
However, thanks to the amazing work by Carnegie Mellon's CTAT team, there are tools that can help you create and run cognitive models (i.e. tha last two bullets). To learn more about JESS and cognitive modeling in general, consider the following links:
- Jess documentation (you will need this to write the production rules for the cognitive model)
- CTAT documentation (you will need this to create the necessary Jess templates and working memory configuration)
This code is written in Java and requires the JESS rule engine. Due to licensing, source code or the JESS library cannot be included in this repo. However, if you do not have JESS, you can download it from here.
Note: Once you download JESS, the code must be packed in a jar file using the command
$ jar cvf jess.jar /path/to/jess/source
After downloading the project (i.e. CogModAPI), edit the build.xml, and set the path to jess.jar in line 9, i.e.
<property name="jessJarPath" value="/path/to/jess/download/jess.jar"></property>
Thats it, after that the project should be compiling and running without glitches!
The project comes with a cognitive model for solving algebra equations, so testing the project is pretty straightforward: all you have to do is compile and run! The main class is ['/src/engine/RuleEngine.java'], so
$ ant # this will compile the code and fill out the /bin folder with the class files.
$ java -cp ./bin/:/path/to/jess.jar engine/RuleEngine 2x+8=12 #alternatively, type ant -Dequation=2x+8=12 run
where 2x+8=12 is the equation you want to solve (i.e. tha task). The output will be a solution displayed on the provided interface, together with a solution graph that has all the necessary steps to solve the problem. E.g. for the problem 2x+8=12 you should see something like this in the output:
-----Cognitive Model Testing------
Solving equation 2x+8=12...
Solution on Interface:
2x+8 = 12 | subtract 8
2x = 4 | divide 2
x = 2 |
Task is Completed!
Solution Graph:
[2x+8=12]
|.....< subtract : wmeTable_C3R1,subtract 8 ✓ >.....[state1]
[state1]
|.....< subtract-typein : wmeTable_C1R2,2x ✓ >.....[state2]
[state2]
|.....< subtract-typein : wmeTable_C2R2,4 ✓ >.....[state3]
[state3]
|.....< divide : wmeTable_C3R2,divide 2 ✓ >.....[state4]
[state4]
|.....< divide-typein : wmeTable_C1R3,x ✓ >.....[state5]
[state5]
|.....< divide-typein : wmeTable_C2R3,2 ✓ >.....[state6]
[state6]
|.....< done : done,-1 ✓ >.....[state7]
Note that each solution step displays the solution step in the following format: * skill applied: interface element, value, correctness*. So "subtract : wmeTable_C3R1,subtract 8 ✓" actually means that skill "subtract" was applied (i.e. production rule named "subtract"), thus the interface element named "wmeTable_C3R1" was filled with the text "subtract 8" which is correct "✓".
The code is organized in two packages (the key classes are covered here):
- engine package : contains the generic code for the cognitive model engine:
- RuleEngine.java: a wrapper for the JESS rule engine.
- CognitiveTask.java: an abstract class holding all the info to perfrom a cognitive task, e.g. the cognitive model, the interface (in a String representation).
- SolutionGraph.java: a class to hold the solution (i.e. how the cognitive model performed on the given task).
- several utility classes to make our life easier when dealing with Jess).
- algebraDomain package : contains the domain dependent code for a cognitive model that solves algebraic equations:
- AlgebraTask: implementation of a CognitiveTask specific for the algebra domain
- MathUtilities.java: a collection of low level domain dependent methods.
- algebraDomain.userDef package: high level domain depedent classes used by the cognitive model.
- wmeTypes.clp: the working memory templates corresponding to the interface elements we will be using
- init.wme: the configuration of the working memory elements (e.g. structure, initial values etc)
- productionRules.pr: the actual cognitive model, a list of production rules describing how to solve equations
To begin, you must have the cognitive model. Rule based cognitive models consist of a production rule file (i.e. productionRules.pr), and the working memory configuration (i.e. wmeTypes.clp and init.wme). These latter files define the necessary structures for the domain under study, and can be generated either by hand (if you are experienced), or automatically using [CTAT] (http://ctat.pact.cs.cmu.edu/docs/ctat_2_6/) (shoot me an email at [email protected] if you have trouble doing so). Withing the current project, you will find working memory files representing a 3x3 matrix structure, which should be sufficient for several tasks involving 9 textboxes.
Once you have the production rules and working memory files, you need to instantiate a ['CognitiveTask'] for your domain.
public class MyTask extends CognitiveTask{
public MyTask(String taskName) {
super(taskName);
}
@Override
public HashMap<String, String> computeInitialWMValuesFromTaskName(){
/* Enter here code that breaks down the task name and returns a hash containing [what workign memory element should be updated]->[updating value].
E.g. for algebra and task name 3x+4=5, we would return [wmeTable_C1R1]->3x+4, [wmeTable_C1R1]->5 */
}
@Override
public String getInitialFactsFile() {
/*return the init.wme filename
e.g. return "/somepath/init.wme";
*/
}
@Override
public String getWMETypesFile() {
/*return the wmeTypes.clp filename
e.g. return "/somepath/wmeTypes.clp";
*/
}
@Override
public String getProductionRulesFile() {
/*return the production rules filename
e.g. return "/somepath/productionRules.pr";
*/
}
@Override
public String getJessUserDefPackageFile() {
/*return the name of the package containing the user defined functions (that are used in the production rules).
e.g. return "algebraDomain.userDef";
*/
}
@Override
public String getTaskInterfaceLayout() {
/*Return a String representation of the interface, show we display the in a meaningfull way. Note that $ behind the name of the workign memroy element means this element is expected to be part of the initial state (that defines the problem).
e.g. return "[$wmeTable_C1R1] = [$wmeTable_C2R1] | [wmeTable_C3R1]\n"
+ "[wmeTable_C1R2] = [wmeTable_C2R2] | [wmeTable_C3R2]\n"
+ "[wmeTable_C1R3] = [wmeTable_C2R3] | [wmeTable_C3R3]\n"
+ "[<doneButton>]";
*/
}
Once this class is implemented, you are done! To test the cognitive model, all you have to do is create an instance of your task, an pass it through an instance of your rule engine, e.g. for the Algebra domain:
AlgebraTask equationProblem=new AlgebraTask("6=4x+8");
RuleEngine algebraSolver=new RuleEngine(equationProblem);
SolutionGraph solution=algebraSolver.runEngine();The solution graph now contains the solution. To display the solution you may either print the solution graph directly, or you may display the solution on the interface:
solution.printGraph(); /*print the solution graph */
System.out.println(equationProblem.getTaskInterface().displaySolution(solution)); /*print the solution on the interface*/To test the cognitive model on another problem, you have to update the existing AlgebraTask with the the problem (or create a new Algebra Task), and reset the rule engine:
equationProblem.updateTaskName("2x=6");
algebraSolver.resetRuleEngine(equationProblem);
solution=algebraSolver.runEngine();This project is licensed under the MIT License - see the LICENSE.md file for details