Solve the Zebra puzzle with miniKanren

The Zebra Puzzle, originally published in Life International in 1962, is a must-have example for logic programming languages. This repository proposes a solution logpy, kanren's ancestor, but does not seem to be working with kanren. In this blog post I will propose a kanren solution, and show how we can make the solution much more readable than other languages' using dataclasses.

I reproduce the puzzle here so you can easily refer to it:

  1. There are five houses.
  2. The Englishman lives in the red house.
  3. The Spaniard owns the dog.
  4. Coffee is drunk in the green house.
  5. The Ukrainian drinks tea.
  6. The green house is immediately to the right of the ivory house.
  7. The Old Gold smoker owns snails.
  8. Kools are smoked in the yellow house.
  9. Milk is drunk in the middle house.
  10. The Norwegian lives in the first house.
  11. The man who smokes Chesterfields lives in the house next to the man with the fox.
  12. Kools are smoked in the house next to the house where the horse is kept.
  13. The Lucky Strike smoker drinks orange juice.
  14. The Japanese smokes Parliaments.
  15. The Norwegian lives next to the blue house.

Now, who drinks water? Who owns the zebra?

In the interest of clarity, it must be added that each of the five houses is painted a different color, and their inhabitants are of different national extractions, own different pets, drink different beverages and smoke different brands of American cigarets [sic]. One other thing: in statement 6, right means your right.


Houses and what's inside them

The first line states that there are 5 houses. We will represent each house with a logic variable Var; the houses are stored in a tuple where the first element represents the leftmost house and the last element the rightmost house. Each house has 4 characteristics: its color, the nationality of the inhabitant, the kind of drink they're having, the kind of animal they own, and the brand of cigarettes they smoke. Given the structure of the problem, dataclasses where fields default to a new Var seems particularly adapted. We can use unification's @unifiable decorator to make the dataclass "unifiable" so we can use it with kanren:

from dataclasses import dataclass, field
from kanren import var, vars
from unification import unifiable

class House():
    nationality: str = field(default_factory=var)
    drink: str = field(default_factory=var)
    animal: str = field(default_factory=var)
    cigarettes: str = field(default_factory=var)
    color: str = field(default_factory=var)

houses = vars(5)

The second statement translates to the following miniKanren goal:

from kanren import membero

membero(House("Englishman", color="red"), houses)

Or, in plain english, "The house that is red and where the Englishman lives is one of the houses". Statements 2, 3, 4, 7, 13, 14 can be translated similarly. Don't forget to speficy that there's a house where someone drinks water and one where someone owns a zebra!

membero(House(drink="water"), houses)
membero(House(animal="zebra"), houses)

Houses' location

Some of the statements are about the houses' location. In particular, statement 6 is aboout a house that is located to the right of another one. We thus need to define a goal that expresses that a house is to the right of another:

def righto(right, left, houses):
    """Express that `right` is on the right of `left` among all the houses."""
    neighbors = tuple(zip(houses[:-1], houses[1:]))
    return membero((left, right), neighbors)

Statements 11, 12, 15 are about houses that are located next to each other. The corresponding goal is easily expressed using the previously-defined righto:

from kanren import conde

def nexto(a, b, houses):
    """Express that `a` and `b` are next to each other."""
    return conde([righto(a, b, houses)], [righto(b, a, houses)])

Statement 10 is about the first house, and statement 9 is about the middle house. Remember that our houses tuple is ordered to these are easily expressed as:

from kanren import eq

eq(House(drink="milk"), houses[2])
eq(House("Norwegian"), houses[0])

The solution

Now that we have defined the concepts and objects we need, we can write the full solution:

from typing import Union
from dataclasses import dataclass, field

from kanren import eq, conde, lall, membero, run
from unification import unifiable, var, vars, Var

class House():
    nationality: Union[str, Var] = field(default_factory=var)
    drink: Union[str, Var] = field(default_factory=var)
    animal: Union[str, Var] = field(default_factory=var)
    cigarettes: Union[str, Var] = field(default_factory=var)
    color: Union[str, Var] = field(default_factory=var)

def righto(right, left, houses):
    """Express that `right` is on the right of `left` among all the houses."""
    neighbors = tuple(zip(houses[:-1], houses[1:]))
    return membero((left, right), neighbors)

def nexto(a, b, houses):
    """Express that `a` and `b` are next to each other."""
    return conde([righto(a, b, houses)], [righto(b, a, houses)])

# And now for the riddle
houses = vars(5)
goals = lall(
    membero(House("Englishman", color="red"), houses),
    membero(House("Spaniard", animal="dog"), houses),
    membero(House(drink="coffee", color="green"), houses),
    membero(House("Ukrainian", drink="tea"), houses),
    righto(House(color="green"), House(color="ivory"), houses),
    membero(House(animal="snails", cigarettes="Old Gold"), houses),
    membero(House(color="yellow", cigarettes="Kools"), houses),
    eq(House(drink="milk"), houses[2]),
    eq(House("Norwegian"), houses[0]),
    nexto(House(cigarettes="Chesterfields"), House(animal="fox"), houses),
    nexto(House(cigarettes="Kools"), House(animal="horse"), houses),
    membero(House(drink="orange juice", cigarettes="Lucky Strike"), houses),
    membero(House("Japanese", cigarettes="Parliaments"), houses),
    nexto(House("Norwegian"), House(color="blue"), houses),
    membero(House(drink="water"), houses),
    membero(House(animal="zebra"), houses),

results = run(0, houses, goals)
([House(nationality='Norwegian', drink='water', animal='fox', cigarettes='Kools', color='yellow'), House(nationality='Ukrainian', drink='tea', animal='horse', cigarettes='Chesterfields', color='blue'), House(nationality='Englishman', drink='milk', animal='snails', cigarettes='Old Gold', color='red'), House(nationality='Spaniard', drink='orange juice', animal='dog', cigarettes='Lucky Strike', color='ivory'), House(nationality='Japanese', drink='coffee', animal='zebra', cigarettes='Parliaments', color='green')],)

The Norwegian drinks water and the Japanese owns the zebra!


miniKanren's python implementation, kanren, allowed us to provide a very intuitive an easy-to-read solution to the Zebra puzzle. Being able to use python's data structures for relational programming goes a long way making miniKanren user friendly, and I hope this convinced you like it did convince me!

In a next post we will see how you can make your own objects "unifiable" and thus use relational programming with an existing codebase. This is already what is happening in aesara, where Ops and Variables have been made unifiable so we can do some pattern matching on Aesara graphs.