pyDatalog: python的逻辑编程引擎【五：与“知识图谱”的交互】

#coding=utf-8
# 与RDF的交互
from pyDatalog import pyDatalog
from pyDatalog.pyDatalog import assert_fact, retract_fact, load, ask
import rdflib
from rdflib import Literal
pyDatalog.create_terms('X,Y,R,relation')
prefix0 = "http://www.blmoistawinde.org/example1#" # URI的统一前缀
abbr = lambda x: x[len(prefix0):]                  # 取URI的缩写，为了展示的简洁
verbose = lambda x: prefix0+x                      # 恢复缩写为URI的全称

g = rdflib.Graph()
g.parse("father_son.ttl", format="turtle")
print("original data:")
for subj, pred, obj in g:          #从RDF取出三元组
    print(abbr(subj),abbr(pred), abbr(obj))
    assert_fact("relation",abbr(subj),abbr(pred), abbr(obj))     #加入Datalog数据库
load("relation(X,'GrandfatherOf',Z) <= relation(X,'FatherOf',Y) & relation(Y,'FatherOf',Z)")

original data:
ZhongShanJingWang IsA ZongQin
LiuBei FatherOf LiuShan

# @prefix : <http://www.blmoistawinde.org/example1#>.

# :ZhongShanJingWang :IsA :ZongQin .

# :LiuBei :FatherOf :LiuShan.

print(relation(X,R,Y))

X                 | R        | Y      
------------------|----------|--------
ZhongShanJingWang | IsA      | ZongQin
LiuBei            | FatherOf | LiuShan

load("+ relation('ZhongShanJingWang','FatherOf','LiuBei')")
load("relation(Y,'IsA','ZongQin') <= relation(X,'IsA','ZongQin') & relation(X,'FatherOf',Y)")
print(relation(X,R,Y))

X                 | R             | Y      
------------------|---------------|--------
ZhongShanJingWang | IsA           | ZongQin
LiuBei            | FatherOf      | LiuShan
ZhongShanJingWang | GrandfatherOf | LiuShan
ZhongShanJingWang | FatherOf      | LiuBei 
LiuShan           | IsA           | ZongQin
LiuBei            | IsA           | ZongQin

# Datalog数据库写入RDF
g2 = rdflib.Graph()
for i in range(len(X.data)):
    subj, pred, obj = verbose(X.data[i]),verbose(R.data[i]),verbose(Y.data[i])
    record = (Literal(subj),Literal(pred),Literal(obj))
    g2.add(record)
print("saved data:")
for subj, pred, obj in g2:          #从RDF取出三元组
    print(abbr(subj),abbr(pred), abbr(obj))
# 用这些语句就可以保存到文件
# str0 = str(g2.serialize(format='turtle')) 
# open("someFile.ttl","w").write(str0)

saved data:
ZhongShanJingWang FatherOf LiuBei
LiuBei IsA ZongQin
LiuBei FatherOf LiuShan
ZhongShanJingWang GrandfatherOf LiuShan
LiuShan IsA ZongQin
ZhongShanJingWang IsA ZongQin

# 再融入networkx进行可视化
import matplotlib.pyplot as plt
import networkx as nx
% matplotlib inline
relation(X,R,Y)
g_nx = nx.MultiDiGraph()
labels = {}
for i in range(len(X.data)):
    subj, pred, obj = X.data[i],R.data[i],Y.data[i]
    g_nx.add_edge(subj,obj)
    labels[(subj,obj)] = pred      #其实可能有多个关系，但这边先只取一个
pos=nx.spring_layout(g_nx)
pos['ZongQin'] = (pos['LiuBei']+pos['LiuShan']+pos['ZhongShanJingWang'])/3.0
nx.draw_networkx_nodes(g_nx, pos, node_size=300)
nx.draw_networkx_edges(g_nx,pos,width=6)
nx.draw_networkx_labels(g_nx,pos,font_size=20,font_family='sans-serif')
nx.draw_networkx_edge_labels(g_nx, pos, labels , font_size=20, font_family='sans-serif')
plt.show()

load("-relation('ZhongShanJingWang','FatherOf','LiuBei')")
print(relation(X,R,Y))
new_relations = [(X.data[i],R.data[i],Y.data[i]) for i in range(len(X.data))]
for subj,pred,obj in g2:
    subj,pred,obj = abbr(subj),abbr(pred), abbr(obj)
    if not (subj,pred,obj) in new_relations:
        print("Fact: %s,%s,%s overturned!"%(subj,pred,obj))

X                 | R        | Y      
------------------|----------|--------
ZhongShanJingWang | IsA      | ZongQin
LiuBei            | FatherOf | LiuShan
Fact: ZhongShanJingWang,FatherOf,LiuBei overturned!
Fact: LiuBei,IsA,ZongQin overturned!
Fact: ZhongShanJingWang,GrandfatherOf,LiuShan overturned!
Fact: LiuShan,IsA,ZongQin overturned!

# 加入更复杂的逻辑，进行一致性的检验，在“加法”后也能“消除”不一致的关系
# 1.互不相交的关系同时成立是不可能的
pyDatalog.clear()
load("relation(X,'Impossible',Y) <= relation(X,R1,Y) & relation(X,R2,Y) & relation(R1,'propertyDisjointWith',R2)")
load("+ relation('FatherOf','propertyDisjointWith','MotherOf')")
load("+ relation('xiaoming','FatherOf','xianghong')")
load("+ relation('xiaoming','MotherOf','xianghong')")
print(relation(X,'Impossible',Y))
for i in range(len(X.data)):               #如果出现不可能的关系，就把相应的两个对象之间的关系都删除
    x0,y0 = X.data[i], Y.data[i]
    relation(x0,R,y0)
    for r0 in R.data:
        retract_fact('relation',x0,r0,y0)
        print("retract_fact:(%s, %s, %s)" % (x0,r0,y0))

X        | Y        
---------|----------
xiaoming | xianghong
retract_fact:(xiaoming, Impossible, xianghong)
retract_fact:(xiaoming, FatherOf, xianghong)
retract_fact:(xiaoming, MotherOf, xianghong)

pyDatalog.clear()
# 2.同一个类不可能同时属于两个不相交的类
load("relation(X,'subClassOf','Nothing') <= relation(X,'subClassOf',C1) & relation(X,'subClassOf',C2) & relation(C1,'disjointWith',C2)")
load("+ relation('GoodPerson','disjointWith','BadPerson')")
load("+ relation('xiaoming','subClassOf','GoodPerson')")
load("+ relation('xiaoming','subClassOf','BadPerson')")
print(relation(X,'subClassOf','Nothing'))
for i in range(len(X.data)):               #如果出现不可能的类，就把相应的类和与其相关的关系都删除
    x0 = X.data[i]
    relation(x0,R,Y)
    for j in range(len(R.data)):
        r0,y0 = R.data[j], Y.data[j]
        retract_fact('relation',x0,r0,y0)
        print("retract_fact:(%s, %s, %s)" % (x0,r0,y0))

X       
--------
xiaoming
retract_fact:(xiaoming, subClassOf, Nothing)
retract_fact:(xiaoming, subClassOf, GoodPerson)
retract_fact:(xiaoming, subClassOf, BadPerson)