Logic programming in R

The goal of logician is to do logic programming inspired by datalog/prolog in R. It is written in R without any third-party package dependencies. It targets interactive use and smaller instances of logical programs.

Non-goal: be fully prolog compatible and super fast.

Side-goal: experiment and have fun.

Installation

You can install the released version of logician from CRAN with:

install.packages("logician")

Or the current github version from here:

remotes::install_github("dirkschumacher/logician")

General principle

The general idea is to query a database of facts and rules and ask questions. logician tries prove that your query is either true or false. Since you can use variables, there might be multiple assignments to variables that make your query true. logician will return these one by one as an iterator.

This is all work in progress and still a bit hacky, but usable.

Example

library(logician)

A graph example

Here we define a database with two types of elements:

1. a fact for each tuple of nodes that are directly connected
2. a rule that determines if there exists a path between two nodes.
   • A path between A and B exists if A and B are connected OR
   • if A is connected to an intermediate node Z and there exists a path from Z to B.

database <- logician_database(
  connected(berlin, hamburg),
  connected(hamburg, chicago),
  connected(chicago, london),
  connected(aachen, berlin),
  connected(chicago, portland),
  connected(portland, munich),
  path(A, B) := connected(A, B),
  path(A, B) := connected(A, Z) && path(Z, B)
)

iter <- logician_query(database, path(berlin, hamburg))
iter$next_value()
#> TRUE

iter <- logician_query(database, path(berlin, munich))
iter$next_value()
#> TRUE

At last let’s find all nodes berlin is connected to.

iter <- logician_query(database, path(berlin, X))
iter$next_value()
#> TRUE
#> X = hamburg.
iter$next_value()
#> TRUE
#> X = chicago.
iter$next_value()
#> TRUE
#> X = london.
iter$next_value()
#> TRUE
#> X = portland.
iter$next_value()
#> TRUE
#> X = munich.
iter$next_value()
#> FALSE

Embracing the host

Unlike prolog, we run on a host language and can integrate R into the evaluation of rules. The only requirement is that the R expression returns a length 1 logical and does not depend on anything outside the globalenv. If that is good idea remains to be seen :) One downside that the list of possible results could be infinite.

Any expression in the r clause will be treated as an R expression that is not unified as the usual clauses. All variables need to be bound before it can be used.

database <- logician_database(
    number(1),
    number(2),
    number(3),
    number(4),
    number(5),
    sum(A, B) := number(A) && number(B) && r(A + B > 3)
)

iter <- logician_query(database, sum(1, B))
iter$next_value()
#> TRUE
#> B = 3.
iter$next_value()
#> TRUE
#> B = 4.
iter$next_value()
#> TRUE
#> B = 5.
iter$next_value()
#> FALSE

Database Example

You could also think of the database as a real database. Where each fact is a row of a specific table. rules are additional logical structures of your data. Then you can use logical programming instead of SQL to query your data.

At the moment this is not as practical though as it could be.

database <- logician_database(
  # the employee table aka relation
  employee(bart),
  employee(homer),
  employee(marge),
  employee(maggie),
  employee(lisa),
  
  # the payment table
  salary(bart, 100),
  salary(homer, 100),
  salary(marge, 120),
  salary(maggie, 140),
  salary(lisa, 180),
  
  # reporting hierarchy
  manages(lisa, maggie),
  manages(lisa, marge),
  manages(marge, homer),
  manages(marge, bart),
  
  # direct and indirect reports
  reports_to(A, B) := manages(B, A),
  reports_to(A, B) := manages(B, X) && reports_to(A, X),
  
  # a salary query
  makes_more(A, X) := salary(A, Y) && r(Y > X + pi)
  # using pi here to show that certain R symbols from the globalenv
  # can be used.
)

# who reports to lisa?
iter <- logician_query(database, reports_to(A, lisa))
iter$next_value()
#> TRUE
#> A = maggie.
iter$next_value()
#> TRUE
#> A = marge.
iter$next_value()
#> TRUE
#> A = homer.
iter$next_value()
#> TRUE
#> A = bart.
iter$next_value()
#> FALSE

# and who makes more than 140 + pi?
iter <- logician_query(database, makes_more(A, 140))
iter$next_value()
#> TRUE
#> A = lisa.
iter$next_value()
#> FALSE

And instead of this manual database, you generate a database with real world data. Though the query system might slow for large amount of at this point.

API

The API has two components: one aimed at interactive use and another one when you want to program with it (e.g. embed it into another package).

• logician_database a helper function to construct a database with the help of non-standard evaluation. A database is just a list of facts and rules that you can also construct yourself using helper functions such as fact, rule, atom, char, clause, variable, int and r_expr.

• logician_query main query function for interactive use. You can use head on an iterator to return the first n results (if they exists).

• logician_query_ same as logician_query but expects a clause or r_expr.

Datatypes and Terminology

• A database contains facts and rules.
• Each fact is a clause.
• A clause has a name and a list of arguments.
• An argument can either be an atom, int, char or var.
• Per convention, variables start with a capital letter, atoms with a lower case letter.
• Each rule has a head and a body.
• A head is a clause.
• A body is a list of clauses or R expressions.

Current Limitations

• Still experimental, so there might be bugs or undefined behavior. API likely will have breaking changes in the future.
• logician_query supports only one clause at the moment. If you want to query for multiple clauses at one, create a rule for that query.
• No list support … yet.
• No cuts. Might never be supported.
• No higher order terms. Though most likely I will not support them.

Test Coverage

covr::package_coverage()
#> logician Coverage: 98.08%
#> R/unification.R: 96.30%
#> R/database.R: 97.50%
#> R/types.R: 97.67%
#> R/query.R: 98.86%