Target only a subset of input in rules

[pierrethiriet]

I am currently moving from R to Julia. I am trying to test the Fuzzy.jl package and I have a small issue with the rule definition.
Is it possible to target only a subset of input in a rule?
In the following example, I a trying to reproduce the classical example of restaurant tip :

Rules:

• IF the service was excellent or the food quality was delicious, THEN the tip will be generous.
• IF the service was good, THEN the tip will be average.
• IF the service was poor and the food quality was poor THEN the tip will be cheap.

service = Dict( "poor"      => TriangularMF(-1, 0, 1.5), 
                "good"      => TriangularMF(3.5, 5, 6.5),
                "excellent" => TriangularMF(8.5, 10, 11.5))

food = Dict("rancid"    => TrapezoidalMF(-2, 0, 2, 4),
            "delicious" => TrapezoidalMF(7, 9, 11, 13))

inputs = [service, food]

tip  = Dict("cheap"     => TriangularMF(2.5, 5, 7.5),
            "average"   => TriangularMF(10, 12.5, 15), 
            "generous"  => TriangularMF(17.5, 20, 22.5))

rule1 = Rule(["poor", "rancid"], "cheap", "MAX")

# How to not include food input ?
rule2 = Rule(["good", ""], "average")

rule3 = Rule(["excellent", "delicious"], "generous", "MAX")
rules = [rule1, rule2, rule3]

fis_tips = FISMamdani(inputs, tip, rules)

So is there a way to avoid including food input for the rule2 ?
Thanks

[phelipe]

Hi!
At the moment it is not possible to do this, but it is a good functionality and I think it is not hard to implement. I will try do this next weekend and report here the results.

[pierrethiriet]

Thanks a lot

[phelipe]

Hi!

Can you test the subset_rule branch? I did a modification and did the model presented in https://iopscience.iop.org/article/10.1088/1742-6596/1830/1/012008/pdf

service = Dict( "poor"      => TrapezoidalMF(0.0, 0.0, 2.0, 4.0), 
                "good"      => TriangularMF(3., 5., 7.),
                "excellent" => TrapezoidalMF(6., 8., 10., 10.))

food = Dict("rancid"    => TrapezoidalMF(0., 0., 3., 6.),
            "delicious" => TrapezoidalMF(4., 7., 10., 10.))

inputs = [service, food]

tip  = Dict("cheap"     => TrapezoidalMF(10., 10., 20., 30.),
            "average"   => TriangularMF(20., 30., 40.), 
            "generous"  => TrapezoidalMF(30., 40, 50., 50.))

rule1 = Rule(["poor", "rancid"], "cheap", "MAX")
rule2 = Rule(["good", ""], "average")
rule3 = Rule(["excellent", "delicious"], "generous", "MAX")
rules = [rule1, rule2, rule3]

fis_tips = FISMamdani(inputs, tip, rules)

in_vals = [7., 8.]

println(eval_fis(fis_tips, in_vals) )

[pierrethiriet]

Thanks a lot,
The new branch works well with the proposed example.
I just have one cometic comment. With more input variables, rules can become complex to follow. For this reason, I use a named tuple in place of the input vector (close to the R Sets package) :

# Convert named tuple to array
inputs_names = [:service, :food]
r(val) = [haskey(val, i) ? getindex(val, i) : "" for i in inputs_names]

rule1 = Rule(r((service="poor", food = "rancid")), "cheap", "MAX")
rule2 = Rule(r((service="good",)), "average")
rule3 = Rule(r((service="excellent", food = "delicious")), "generous", "MAX")
rules = [rule1, rule2, rule3]

I don't know if this syntax could be relevant for others?

[phelipe]

@pierrethiriet I created a new issue related to this modification in syntax, for now I will close this issue and merge the changes with master.