frules 0.1.2

Linguistic variables and expressions

Expression is a core concept in frules. Expression class represents subrange of linguistic variable in fuzzy logic.

Variables in classical math take numerical values. in fuzzy logic, the linguistic variables are non-numeric and are described with expressions. Expressions map continuous variable like nemerical temperature to its linguistic counterpart. For example temperature can be described as cold, warm or hot. There is no strict boundary between cold and warm - this is why this expressions are fuzzy.

To create new expression we use function that takes numerical value of contiunous variable and returns truth value. Truth value ranges between 0 and 1 - it’s a degree of membership of continous value to that linguistic variable.

from frules.expressions import Expression
#We know that anything over 50 degrees is hot and below 40 is't hot
hot = Expression(lambda x: min(1, max((x - 40) / 10., 0)))

This ugly lambda is representation of some fuzzy set. If we take a look how it behaves, we’ll see that it in fact returns 1 for anything over 50, 0 for anything below 40 and some linear values between 40 and 50:

>>> map(lambda x: {x: min(1, max((x - 40) / 10., 0))}, xrange(35, 55, 2))
[{35: 0}, {37: 0}, {39: 0}, {41: 0.1}, {43: 0.3}, {45: 0.5}, {47: 0.7}, {49: 0.9}, {51: 1}, {53: 1}

Using a lot of lambdas in practice makes your code a mess. Fuzzy expressions described this way are additionally hard to write because of some value assertions they must satisfy.

This is why we ancapsulate don’t use raw functions and encapsulate them with expressions. Moreover frules provides a bunch of helpers that eases definition of new expressions. Example of full set of expressions for temperature variable could look this way:

from frules.expressions import Expression as E
from frules.expressions import ltrapezoid, trapezoid, rtrapezoid

cold = E(ltrapezoid(10, 20), "cold")        # anything below 10, more is fuzzy
warm = E(trapezoid(10, 20, 30, 35), "warm") # anything between 20 and 30
hot = E(rtrapezoid(30, 35), "hot")          # anything over 35, less is fuzzy

Expressions can be reused/mixed using logical operators:

cold_or_hot = cold || warm
not_hot = !hot

Optional names will be helpful when we start to work with fuzzy rules.

Fuzzy rules

Although expressions define linguistic variables, they aren’t strictly bound to any variable. They are rather the adjectives we use to describe something and their meaning depends strictly on context. Both person and data could be big but this particular adjective has slighlty different meaning in each case.

Rule objects bounds continous variable with expressions. Rules also can also be evaluated to see how true they are for given continous input.

>>> from frules.rules import Rule
>>> is_hot = Rule(temperature=hot)
>>> is_hot.eval(temperature=5)
0.8

Rules can be mixed using logical operators (& and |) to create more sophisticated rules that allow fuzzy reasoning:

from frules.expressions import Expression as E
from frules.rules import Rule as R
from frules.expressions import ltrapezoid, trapezoid, rtrapezoid

# car age expressions
old = E(ltrapezoid(2001, 2008), "old")
new = E(rtrapezoid(2013, 2014), "new")
not_so_old = - (old & new)

# power expressions
strong = E(rtrapezoid(50, 100), "strong")
weak = E(ltrapezoid(50, 100), "weak")

# price expression
expensive = E(rtrapezoid(25000, 30000), "expensive")
cheap = - expensive

# yes expression
yes = E(lambda yes: float(yes), "yes") # converts bool to float


# rules
is_attractive = R(production_year=not_so_old) & R(horsepower=strong)
should_buy = is_attractive & R(price=cheap)

Having such set of rules we can do some reasoning:

>>> should_buy
(((age = !(old & new) & horsepower = strong) & !None = None) & cost = !expensive)
>>> should_buy.eval(horsepower=70, production_year=2012, price=15000)
0.4
>>>
>>> candidates = {
...     "car1": {"horsepower": 70, "production_year": 2012, "price": 15000},
...     "car2": {"horsepower": 150, "production_year": 2010, "price": 30000},
...     "car3": {"horsepower": 90, "production_year": 2014, "price": 10000},
...     "car4": {"horsepower": 85, "production_year": 2009, "price": 35000},
... }
>>> max(candidates.iteritems(), key=lambda (key, inputs): is_hot.eval(**inputs))
('car3', {'horsepower': 90, 'price': 10000, 'production_year': 2014})