syntreenet 1.0.0b4

Example Usage

Let’s start with a simple grammar with which we can build triples consisting on a subject, a predicate, and an object, with 2 predicates, ∈ and ⊆, thus producing classifications of things. To make this example simpler, we will use the ASCII string “element-of” in place of unicode ∈, and “subset-of” in place of ⊆.

The grammar for this might be something like:

fact        = word ws pred ws word
pred        = element / subset
element     = "element-of"
subset      = "subset-of"
word        = ~"[a-z0-9]+"
ws          = ~"\s+"

With this, we can have “facts” such as:

a element-of b
b subset-of c
c subset-of d

On top of this language, we might want a logic where, if you add the previous 3 facts to a knowledge base, you would also have that:

a element-of c
a element-of d
b subset-of d

For this, we need a logic in which variables range over the “word” productions. So we modify the grammar substituting the “word” rule with:

word        = v_word / __var__
v_word      = ~"[a-z0-9]+"

With this grammar we can now build a knowledge base, and add rules appropriate for our purposes:

>>> from syntreenet import KnowledgeBase

>>> grammar = """
>>>    fact        = word ws pred ws word
>>>    pred        = element / subset
>>>    element     = "element-of"
>>>    subset      = "subset-of"
>>>    word        = v_word / __var__
>>>    v_word      = ~"[a-z0-9]+"
>>>    ws          = ~"\s+"
>>> """

>>> kb = KnowledgeBase(grammar)

>>> kb.tell("X1 element-of X2 ; X2 subset-of X3 -> X1 element-of X3")
>>> kb.tell("X1 subset-of X2 ; X2 subset-of X3 -> X1 subset-of X3")

Now we can add facts to the knowledge base, and query for them or for their conscuences:

>>> kb.tell("a element-of b")
>>> kb.tell("b subset-of c")
>>> kb.tell("c subset-of d")

>>> kb.query("a element-of b")
True
>>> kb.query("a element-of d")
True
>>> kb.query("c element-of d")
False
>>> kb.query("X1 subset-of d")
[{'X1': 'b'}, {'X1': 'c'}]

Goals

Scalability:

Adding new facts or rules is essentially O(1) in the number of rules plus facts already present in the knowledge base. Theoretically, this is due to the fact that the DAGs that hold the data are only ever searched by consulting Python dictionaries. Practically, I am getting a fairly constant value of a couple tenths of a millisecond per fact (this will depend on the complexity of the grammar), up to the capacity of my laptop (totalling around 2 million facts and rules).

Universality:

The “free” in the heading caption is in the sense of a “free object” over the “set of PEGs”: syntreenet knows nothing about the grammar underlying the particular logic it deals with at any particular moment.

Clear and tested code:

The code follows best practices for readability and is tested with 99% coverage including branch analysis.

Detailed Usage

$ pip install syntreenet
$ python
>>> import syntreenet

$ git clone https://git.sr.ht/~enriquepablo/syntreenet
$ cd syntreenet/
$ virtualenv venv
$ source venv/bin/activate
$ python setup.py develop easy_install syntreenet[testing]
$ nose2