paspailleur

An add-on for caspailleur to work with Pattern Structures

A Pattern Structure (D, ⊑) represents a description space D where every two descriptions can be compared by a "less precise" operator ⊑. For example, if D is a set of ngrams then ngram (hello,) is less precise then (hello, world): (hello, ) ⊑ (hello, world), that is every ngram that contains (hello, world) contains (hello,).

Warning

The package is in active development stage. Things can change often.

Implemented Pattern Structures

from paspailleur import pattern_structures as PS

General use

IntervalPS

Every description is a closed interval of real numbers [a,b]. Description [a,b] is less precise than description [c,d] if a<=c, d<=b. For example, description [1.5, 3.14] is less precise than [2, 3], i.e. [1.5, 3.14] ⊑ [2, 3] (yes the notation is counterintuitive here).

d1, d2 = (1.5, 3.14), (2, 3)
ps = PS.IntervalPS()
assert ps.is_less_precise(d1, d2)

SubSetPS

Every description is a set of values. Description A is less precise than description B if A is a subset of B: A ⊆ B. For example description {green, cubic} is less precise than {green, cubic, heavy}.

d1, d2 = {'green', 'cubic'}, {'green', 'cubic', 'heavy'}
ps = PS.SubSetPS()
assert ps.is_less_precise(d1, d2)

SuperSetPS

Every description is a set of values. Description A is less precise than description B if A is a superset of B: A ⊇ B. For example description {green, yellow, red} is less precise than {green, yellow}.

d1, d2 = {'green', 'yellow', 'red'}, {'green', 'yellow'}
ps = PS.SuperSetPS()
assert ps.is_less_precise(d1, d2)

CartesianPS

A pattern structure to combine various independent basic pattern structures in one.

# Combining three previous examples together
d1 = [(1.5, 3.14), {'green', 'cubic'}, {'green', 'yellow', 'red'}]
d2 = [(2, 3), {'green', 'cubic', 'heavy'}, {'green', 'yellow'}]
basic_structures = [PS.IntervalPS(), PS.SubSetPS(), PS.SuperSetPS()]
ps = PS.CartesianPS(basic_structures)
assert ps.is_less_precise(d1, d2)

NLP

NgramPS

Every description is a set of incomparable ngram, i.e. set of incomparable tuple of words.

Ngram A = (a_1, a_2, ..., a_n) is less precise than ngram B = (b_1, b_2, ..., b_m) if A can be embedded into B: i.e. exists i = 1, ..., m-n s.t. A = B[i:i+n]. For example (hello, world) is less precise than (hello, world, !).

Description D_1 = {A_1, A_2, ...} is less precise than description D_2 = {B_1, B_2, ...} if every ngram in D1 is less precise than some ngram in D2.

d1 = {('hello', 'world'), ('!',)}
d2 = {('hello', 'world', '!')}
ps = PS.NgramPS()
assert ps.is_less_precise(d1, d2)

SynonymPS

Every description is a set of words, representing the synonyms of words contained in some text. Description A is less precise than description B if A is a subset of B: A ⊆ B.

d1, d2 = 'hello', 'hello world'
ps = PS.SynonymPS()
pattern1, pattern2 = ps.preprocess_data([d1, d2])
assert ps.is_less_precise(pattern1, pattern2)

print('pattern1:', pattern1)
print('pattern2:', pattern2)

pattern1: {'hello'}
pattern2: {'hello', 'universe'}

AntonymPS

Every description is a set of words, representing the antonyms of words contained in some text. Description A is less precise than description B if A is a subset of B: A ⊆ B.

d1, d2 = 'good', 'good dog'
ps = PS.AntonymPS()
pattern1, pattern2 = ps.preprocess_data([d1, d2])
assert ps.is_less_precise(pattern1, pattern2)

print('pattern1:', pattern1)
print('pattern2:', pattern2)

pattern1: {'evil'}
pattern2: {'evil'}

So, the system does not know any antonym to "dog".

Tabular data

[!INFO] Coming soon

NumberPS

CategoryPS

Graphs

[!INFO] Coming soon

Coming soon

GraphPS

OrderedGraphPS

How to create a custom Pattern Structure

To be described