Migration to jfuzzylite

[Unknown]

I’m trying to migrate from fuzzylogic to jfuzzylite, but the defuzzification results seem very different.

Input values:
signal: 30
distance: 40

Output values:
fuzzylogic: 24.473140
jfuzzylite: 57.058

What I’m doing wrong?

FCL:

FUNCTION_BLOCK proximity

VAR_INPUT
    signal : REAL;
    distance : REAL;
END_VAR

VAR_OUTPUT
    proximity : REAL;
END_VAR

FUZZIFY signal
    TERM strong := (0,1) (30,1) (65, 0);
    TERM decent := (25, 0) (60,1) (90, 0);
    TERM weak := (80, 0) (120, 1);
END_FUZZIFY

FUZZIFY distance
    TERM low := (0,1) (20,0);
    TERM medium := (15, 0) (30,1) (70, 0);
    TERM high := (60, 0) (200, 1);
END_FUZZIFY

DEFUZZIFY proximity
    TERM immediate := (0, 1) (30, 0);
    TERM near := (20, 0) (40, 1) (80, 0);
    TERM far := (70, 0) (100, 1);
        
        
    METHOD : COG;        // Use 'Center Of Gravity' defuzzification method
    DEFAULT := 0;        // Default value is 0 (if no rule activates defuzzifier)
END_DEFUZZIFY

RULEBLOCK first     
    AND : MIN;
        
    ACCU : MAX;
    ACT : MIN;
        
    RULE 1 : IF signal IS strong AND distance IS low THEN proximity IS immediate;
    RULE 2 : IF signal IS strong AND distance IS medium THEN proximity IS immediate;
    RULE 3 : IF signal IS strong AND distance IS high THEN proximity IS immediate;
    RULE 4 : IF signal IS decent AND distance IS low THEN proximity IS immediate;
    RULE 5 : IF signal IS decent AND distance IS medium THEN proximity IS near;
    RULE 6 : IF signal IS decent AND distance IS high THEN proximity IS near;
    RULE 7 : IF signal IS weak AND distance IS low THEN proximity IS immediate;
    RULE 8 : IF signal IS weak AND distance IS medium THEN proximity IS near;
    RULE 9 : IF signal IS weak AND distance IS high THEN proximity IS far;
END_RULEBLOCK

END_FUNCTION_BLOCK

jfuzzylite:

Engine engine = new Engine();
engine.setName("Proximity");
engine.setDescription("");
        
InputVariable signal = new InputVariable();
signal.setName("signal");
signal.setDescription("");
signal.setEnabled(true);
signal.setRange(30.0, 120.0);
signal.setLockValueInRange(true);
signal.addTerm(new Trapezoid("strong", 0.0, 0.0, 30.0, 65.0));
signal.addTerm(new Triangle("decent", 25.0, 60.0, 90.0));
signal.addTerm(new Ramp("weak", 80.0, 120.0));
engine.addInputVariable(signal);
        
InputVariable distance = new InputVariable();
distance.setName("distance");
distance.setDescription("");
distance.setEnabled(true);
distance.setRange(0.0, 200.0);
distance.setLockValueInRange(true);
distance.addTerm(new Ramp("low", 0.0, 20.0));
distance.addTerm(new Triangle("medium", 15.0, 30.0, 70.0));
distance.addTerm(new Ramp("high", 60.0, 200.0));
engine.addInputVariable(distance);
    
OutputVariable proximity = new OutputVariable();
proximity.setName("proximity");
proximity.setDescription("");
proximity.setEnabled(true);
proximity.setRange(0.0, 100.0);
proximity.setLockValueInRange(false);
proximity.setAggregation(new Maximum());
proximity.setDefuzzifier(new Centroid(100));
proximity.setDefaultValue(Double.NaN);
proximity.setLockPreviousValue(false);
proximity.addTerm(new Ramp("immediate", 0.0, 30.0));
proximity.addTerm(new Triangle("near", 20.0, 40.0, 80.0));
proximity.addTerm(new Ramp("far", 70.0, 100.0));
engine.addOutputVariable(proximity);
        
RuleBlock mamdani = new RuleBlock();
mamdani.setName("mamdani");
mamdani.setDescription("");
mamdani.setEnabled(true);
mamdani.setConjunction(new AlgebraicProduct());
mamdani.setDisjunction(new AlgebraicSum());
mamdani.setImplication(new Minimum());
mamdani.setActivation(new General());
mamdani.addRule(Rule.parse("if signal is strong and distance is low then proximity is immediate", engine));
mamdani.addRule(Rule.parse("if signal is strong and distance is medium then proximity is immediate", engine));
mamdani.addRule(Rule.parse("if signal is strong and distance is high then proximity is immediate", engine));
mamdani.addRule(Rule.parse("if signal is decent and distance is low then proximity is immediate", engine));
mamdani.addRule(Rule.parse("if signal is decent and distance is medium then proximity is near", engine));
mamdani.addRule(Rule.parse("if signal is decent and distance is high then proximity is near", engine));
mamdani.addRule(Rule.parse("if signal is weak and distance is low then proximity is immediate", engine));
mamdani.addRule(Rule.parse("if signal is weak and distance is medium then proximity is near", engine));
mamdani.addRule(Rule.parse("if signal is weak and distance is high then proximity is far", engine));
engine.addRuleBlock(mamdani);
        
signal.setValue(30.0);
distance.setValue(40.0);
engine.process();
FuzzyLite.logger().info(String.format(
    "signal.input = %s, distance.input = %s -> proximity.output = %s",
    Op.str(signal.getValue()), Op.str(distance.getValue()), Op.str(proximity.getValue())));

[Unknown]

Solved. The error was in Ramp values:

distance.addTerm(new Ramp("low", 0.0, 20.0));
proximity.addTerm(new Ramp("immediate", 0.0, 30.0));

Should be:

distance.addTerm(new Ramp("low", 20.0, 0.0));
proximity.addTerm(new Ramp("immediate", 30.0, 0.0));

But, I find another issue. I’m exporting this code to R:

RScriptExporter exporter = new RScriptExporter();
exporter.toFile(new File("plot.r"), engine);

When I’m trying to execute, RStudio gives me the following error:

Error: Discrete value supplied to continuous scale
In addition: Warning message:
  Error: Discrete value supplied to continuous scale 

[Juan Rada-Vilela (admin)]

Hi Andrey,

thanks for your post and for using jfuzzylite.

Could you please post the FLL code?

As for the previous question, glad you found the mistake. The FCL engine you posted, the equivalent in FLL is:

Engine: proximity
InputVariable: signal
  enabled: true
  range: 30.000 120.000
  lock-range: false
  term: strong Discrete 0.000 1.000 30.000 1.000 65.000 0.000
  term: decent Discrete 25.000 0.000 60.000 1.000 90.000 0.000
  term: weak Discrete 80.000 0.000 120.000 1.000
InputVariable: distance
  enabled: true
  range: 0.000 200.000
  lock-range: false
  term: low Discrete 0.000 1.000 20.000 0.000
  term: medium Discrete 15.000 0.000 30.000 1.000 70.000 0.000
  term: high Discrete 60.000 0.000 200.000 1.000
OutputVariable: proximity
  enabled: true
  range: 0.000 100.000
  lock-range: false
  aggregation: Maximum
  defuzzifier: Centroid 100
  default: 0.000
  lock-previous: false
  term: immediate Discrete 0.000 1.000 30.000 0.000
  term: near Discrete 20.000 0.000 40.000 1.000 80.000 0.000
  term: far Discrete 70.000 0.000 100.000 1.000
RuleBlock: first
  enabled: true
  conjunction: Minimum
  disjunction: none
  implication: Minimum
  activation: General
  rule: if signal is strong and distance is low then proximity is immediate
  rule: if signal is strong and distance is medium then proximity is immediate
  rule: if signal is strong and distance is high then proximity is immediate
  rule: if signal is decent and distance is low then proximity is immediate
  rule: if signal is decent and distance is medium then proximity is near
  rule: if signal is decent and distance is high then proximity is near
  rule: if signal is weak and distance is low then proximity is immediate
  rule: if signal is weak and distance is medium then proximity is near
  rule: if signal is weak and distance is high then proximity is far

Take a look at the engine:

[Unknown]

Hi Juan,

Here is the generated R code by jfuzzylite.

When you run this code the following error occurs:

Error in textConnection(engine.fld) : object 'engine.fld' not found

So I changed from “engine.fld” to “engine.fldFile” on line 1075. But the original error still persists:

Error: Discrete value supplied to continuous scale
In addition: Warning message:
  Error: Discrete value supplied to continuous scale

[Juan Rada-Vilela (admin)]

Hi,

could you please post the FLL code of your engine?

Thanks.

[Juan Rada-Vilela (admin)]

oh, never mind. just saw it in the R code.

[Juan Rada-Vilela (admin)]

I think the problem you are having is with locales. Take notice that you export values with , as decimal separators. I think R is not handling this well.

Please try changing the locale in java before exporting to something like Locale.setDefault(new Locale("en", "NZ"));, make sure the exported values use . for decimals, and try running the script in R.

You do not need to change anything from the R script.

Cheers.

[Unknown]

It worked! Thank you Juan.

[Author]

Posts