forked from palmagro/gg2vec
-
Notifications
You must be signed in to change notification settings - Fork 0
/
node2vec.py
493 lines (460 loc) · 19.5 KB
/
node2vec.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
from collections import OrderedDict
import json
import pickle as pickle
import os.path
from sklearn import manifold
from bokeh.io import output_notebook
from bokeh.plotting import figure, output_file, show, ColumnDataSource, vplot
from bokeh.models import(
GMapPlot, Range1d, ColumnDataSource, LinearAxis,
PanTool, WheelZoomTool, BoxZoomTool, ResetTool, ResizeTool, BoxSelectTool, HoverTool)
from bokeh.charts import Line
from gensim.models import word2vec
import logging
import random
from py2neo import neo4j
import numpy as np
import scipy as scipy
import sys # sys.setdefaultencoding is cancelled
reload(sys) # to re-enable sys.setdefaultencoding()
import numpy.linalg as la
from bokeh.charts import BoxPlot
from scipy import signal
import math
from overs import *
from aux import *
from visualization import *
from ZODB import DB
from ZODB.FileStorage import FileStorage
from ZODB.PersistentMapping import PersistentMapping
import transaction
from persistent import Persistent
from persistent.dict import PersistentDict
from persistent.list import PersistentList
from copy import *
from joblib import Parallel, delayed
import multiprocessing
sys.setdefaultencoding('utf-8')
class node2vec:
r_deleted = {}
sentences = {}
sentences_array = []
degree = []
r_types = []
n_types = []
r_types_d = []
r_desv = {}
n_types_d = []
m_vectors = []
m_points = []
angle_matrix= []
plotw = 800
ploth = 500
mode = "normal"
def __init__(self,bd,port,user,pss,label,ns,nd,l,m,traversals,iteraciones):
self.nodes = []
self.ndim = nd
self.bd = bd
self.port = port
self.user = user
self.pss = pss
self.label = label
self.ns = ns
self.w_size = l
self.mode = m
self.iteraciones = iteraciones
# Setting up Neo4j DB
neo4j.authenticate("http://localhost:"+str(self.port), self.user, self.pss)
self.graph_db = neo4j.GraphDatabaseService("http://neo4j:"+pss+"@localhost:"+str(self.port)+"/db/data/")
batches = 100
if not os.path.exists("models/" + self.bd +".npy") or not os.path.exists("models/" + self.bd +"l-degree.npy"):
print "Conecting to BD..."
nn = neo4j.CypherQuery(self.graph_db, "match n return count(n) as cuenta1").execute()
self.numnodes = nn[0].cuenta1
self.sentences_array = []
nb = float(self.numnodes/batches)
count = -1
self.degree = []
for i in range(1,int(nb)+1):
count += 1
consulta = "match (n)-[r]-(m) where n."+self.label+" <> '' return n,count(r) as d, n."+self.label+", collect(m."+self.label+") as collect skip "+str(batches*count)+" limit "+str(batches)
cuenta = neo4j.CypherQuery(self.graph_db, consulta).execute()
print "\r"+str(float((i / nb)*100))+ "%"
for cuenta1 in cuenta:
name = cuenta1['n.'+label].replace(" ","_")
context = []
#Extracting context(relations)
for s in cuenta1['collect']:
if type(s) is list:
for x in s:
context.append(str(x).replace(" ","_"))
else:
if s:
context.append(str(s).replace(" ","_"))
#Extracting contexto(properties)
for t in cuenta1['n']:
s = cuenta1['n'][t]
if type(s) is list:
for x in s:
context.append(str(x).replace(" ","_"))
else:
if s:
context.append(str(s).replace(" ","_"))
if len(context) >= l-1 and cuenta1.d is not None:
sentence = context
sentence.insert(0,name)
self.sentences_array.append(sentence)
self.degree.append(cuenta1.d)
np.save("models/" + self.bd , self.sentences_array)
np.save("models/" + self.bd +"l-degree", self.degree)
else:
self.sentences_array = np.load("models/" + self.bd +".npy")
self.degree = np.load("models/" + self.bd +"l-degree.npy")
for s in self.sentences_array:
self.sentences[s[0]]=s[1:]
print "models/" + self.bd +".npy"
def learn(self,m,ts,d,it):
num_cores = multiprocessing.cpu_count()
print "numCores = " + str(num_cores)
self.path = "models/" + self.bd + str(self.ndim) +"d-"+str(self.ns)+"w"+str(self.w_size)+"l"+m
if d:
#el metodo delete_rels elimina las relaciones por las que despues preguntaremos de self.sentences_array antes de entrenar y devuelve el nuevo dump con las relaciones quitadas y una lista de las relaciones quitadas
self.learn(m,0,False,it)
self.get_rels([])
sents,self.r_deleted = delete_rels(self.sentences_array,self.r_types,ts)
self.path = self.path + "del"+str(ts)+"-"
else:
sents = self.sentences_array
self.path = self.path +str(it)+".npy"
print "Learning:" + self.path
print "CCCC!"
if not os.path.exists(self.path):
print "Entra"
entrada = []
results = Parallel(n_jobs=num_cores, backend="threading")(delayed(generate_sample)(self.mode,sents,self.degree,self.w_size,i) for i in range(1,self.ns))
for r in results:
entrada.append(r)
self.w2v = word2vec.Word2Vec(entrada, size=self.ndim, window=self.w_size, min_count=1, workers=num_cores,sg=0)
self.w2v.save(self.path)
print "TERMINO"
else:
self.w2v = word2vec.Word2Vec.load(self.path)
self.get_nodes()
self.get_rels([])
self.delete_props()
def get_rels(self,traversals):
if not os.path.exists("models/" + self.bd+"-trels.p"):
f = open( "models/" + self.bd+"-trels.p", "w" )
consulta = neo4j.CypherQuery(self.graph_db, "match (n)-[r]->(m) return n."+self.label+" as s,m."+self.label+" as t ,r,type(r) as tipo,labels(m) as tipot").execute()
todas = []
for c in consulta:
todas.append([c.s,c.tipo,c.t,c.tipot])
pickle.dump(todas,f)
else:
f = open( "models/" + self.bd+"-trels.p", "r" )
todas = pickle.load(f)
links = dict()
for l in todas:
link = dict()
if l[0] and l[1] and l[2]:
link["tipo"] = l[1]
link["s"] = l[0].replace(" ","_")
link["t"] = l[2].replace(" ","_")
link["tipot"] = l[3][0].replace(" ","_")
if link["s"] in self.w2v and link["t"] in self.w2v:
link["v"] = self.w2v[link["t"]] - self.w2v[link["s"]]
if not link["tipo"] in links:
links[link["tipo"]] = []
links[link["tipo"]].append(link)
self.r_types = links
def r_analysis(self):
print "Relation Types Analysis"
if self.r_types == []:
self.get_rels()
self.m_vectors = {}
for t in self.r_types:
vectors = []
rels = self.r_types[t]
for r in rels:
if (r["s"] in self.w2v) and (r["t"] in self.w2v):
vectors.append(self.w2v[r["t"]] - self.w2v[r["s"]])
vector_medio = np.mean(vectors,axis=0)
self.m_vectors[t] = np.mean(vectors,axis=0)
media = 0
for v in vectors:
media = media + angle(v,vector_medio)
media = media / len(vectors)
self.r_desv[t] = media
print "Mean Vector Angles"
self.angle_matrix= dict()
for i,t in enumerate(self.r_types):
self.angle_matrix[t] = dict()
for j,x in enumerate(self.r_types):
self.angle_matrix[t][x] = angle(self.m_vectors[t],self.m_vectors[x])
if x not in self.angle_matrix:
self.angle_matrix [x]= dict()
self.angle_matrix[x][t] = angle(self.m_vectors[t],self.m_vectors[x])
def get_nodes(self):
if not os.path.exists("models/" + self.bd+"-tnodes.p"):
f = open( "models/" + self.bd+"-tnodes.p", "w" )
consulta = neo4j.CypherQuery(self.graph_db, "match (n) return n."+self.label+" as name,labels(n) as tipos").execute()
nodes = dict()
for node in consulta:
if node.name and node.tipos <> []:
name = node.name.replace(" ","_")
for tipo in node.tipos:
if not tipo in nodes:
nodes[tipo] = []
nodes[tipo].append(name)
self.n_types = nodes
pickle.dump(nodes,f)
else:
f = open( "models/" + self.bd+"-tnodes.p", "r" )
self.n_types = pickle.load(f)
def n_analysis(self):
print "Node Type Analysis"
if self.n_types == []:
self.get_nodes()
self.m_points = dict()
self.n_types_d = dict()
for nt in self.n_types:
points = []
for node in self.n_types[nt]:
if node in self.w2v:
points.append(self.w2v[node])
if len(points) > 0:
punto_medio = [0] * len(points[0])
for p in points:
for idx,d in enumerate(p):
punto_medio[idx] = punto_medio[idx] + d
for idx,d in enumerate(punto_medio):
punto_medio[idx] = punto_medio[idx] / len(points)
if nt not in self.m_points:
self.m_points[nt] = punto_medio
#print "-------------------"+nt+"-------------------"
#print "Number of Nodes: "+ str(len(points))
dev = 0
for p in points:
dev = dev + scipy.spatial.distance.euclidean(punto_medio,p)**2
dev = math.sqrt((dev / len(points)))
#print "Standard Deviation:"+str(dev)
if nt not in self.n_types_d:
self.n_types_d[nt] = dev
#print "Variance:"+str(np.var(points))
#print "Distancia entre los puntos medios"
#for i,t in enumerate(self.m_points):
#for j,x in enumerate(self.m_points):
#if i <> j:
#print t+" vs. "+x
#print scipy.spatial.distance.euclidean(self.m_points[t] , self.m_points[x])
def analysis(self):
self.n_analysis()
self.r_analysis()
def similares(self,nodo,positives,negatives,top_n,filtrado):
#Version nueva: utilizo las estructuras nodes_pos y nodes_type en un knn de scikit
clf = neighbors.KNeighborsClassifier(top_n, "uniform",n_jobs=multiprocessing.cpu_count())
clf.fit(self.nodes_pos, self.nodes_type)
my_list = clf.kneighbors(positives[0],top_n,False)
#Version antigua: usaba word2ec por lo que estaba trabajando con todas las propiedades
my_list = self.w2v.most_similar(positives,negatives,topn=top_n)
result = []
for m in my_list:
if m[0] != nodo:
result.append(m)
return result
def predice(self,nodo,rel,fast,top_n,filtrado):
if not fast:
votos = []
for r in self.r_types[rel]:
other = r["s"]
if(r["s"] == nodo):
other = r["t"]
p2 = neo4j.CypherQuery(self.graph_db, "match (n)-[:"+rel[0]+"]-(m) where n."+label+' = "'+other+'" return m.'+label).execute()
print p2
if len(p2) > 0:
for p in p2:
prop2 = p["m."+label]
prop2 = prop2.replace(" ","_")
other = other.replace(" ","_")
if other in self.w2v and prop2 in self.w2v:
prop1 = self.similares([nodo,other],[prop2])[0][0]
votos.append(prop1)
return max(set(votos), key=votos.count)
if fast:
sim = self.similares(nodo,[self.w2v[nodo]+self.m_vectors[str(rel)]],[],top_n,filtrado)
f = []
for s in sim:
f.append(s[0])
if len(f) > 0:
return f
else:
return ""
def aciertos_rel(self,rel,label,fast,string):
print "jeje"
if not os.path.exists("models/" + self.bd + str(self.ndim) +"d-"+str(self.ns)+"w"+str(self.w_size)+self.mode+"-lpr-"+rel+string+".p"):
print "ta"
parcial = 0
total = 0
cuenta_misc = 0
for d in self.r_deleted[rel]:
print "analizando relacion"
print rel
rs = d["s"]
cuenta_misc += 1
print rs
print rs in self.w2v
print rs in self.sentences
if rs in self.w2v and not '"' in rs:
total = total + 1
nbs = self.predice(rs,label,self.r_types1[rel]["t"],rel,fast)
if d["t"] in nbs:
print "HOLA"
print d["t"]
print nbs.index(d["t"])
parcial += float(1 / float(nbs.index(d["t"])+1 ))
print parcial
print total
if total > 0:
result = float(parcial)/float(total)
else:
result = 0
f = open( "models/" + self.bd + str(self.ndim) +"d-"+str(self.ns)+"w"+str(self.w_size)+self.mode+"-lpr-"+rel+string+".p", "w" )
pickle.dump(result,f)
else:
f = open( "models/" + self.bd + str(self.ndim) +"d-"+str(self.ns)+"w"+str(self.w_size)+self.mode+"-lpr-"+rel+string+".p", "r" )
result = pickle.load(f)
return result
def link_prediction_ratio(self):
ratiosf = {}
for r in self.r_types:
ratiosf[r] = self.aciertos_rel(r,self.label,True)
xname = []
yname = []
alpha = []
color = []
ratio = []
names=[]
for r in self.r_types:
names.append(r)
xname.append(r)
yname.append("Ratio")
alpha.append(ratiosf[r]/100)
ratio.append(ratiosf[r])
color.append('black')
source = ColumnDataSource(
data=dict(
xname=xname,
yname=yname,
colors=color,
alphas=alpha,
ratios=ratio
)
)
p = figure(title="Link Prediction Ratios",
x_axis_location="above", tools="resize,hover,save",
x_range=xname, y_range=["Ratio"])
p.rect('xname', 'yname', 0.9, 0.9, source=source,
color='colors', alpha='alphas', line_color=None)
p.grid.grid_line_color = None
p.axis.axis_line_color = None
p.axis.major_tick_line_color = None
p.axis.major_label_text_font_size = "5pt"
p.axis.major_label_standoff = 0
p.xaxis.major_label_orientation = np.pi/3
hover = p.select(dict(type=HoverTool))
hover.tooltips = OrderedDict([
('link type and method', '@yname, @xname'),
('link prediction ratio', '@ratios'),
])
return p
def ntype(self,n):
for t in self.n_types:
if n in self.n_types[t]:
return t
def connectZODB(self):
print "connnecting"
if not os.path.exists(self.bd+'.fs'):
self.storage = FileStorage(self.bd+'.fs')
self.db = DB(self.storage)
self.connection = self.db.open()
self.root = self.connection.root()
self.root = PersistentDict()
else:
self.storage = FileStorage(self.bd+'.fs')
self.db = DB(self.storage)
self.connection = self.db.open()
self.root = self.connection.root()
def disconnectZODB(self):
print "grabando!"
transaction.commit()
self.connection.close()
self.db.close()
self.storage.close()
#Creating nodes_pos dictionary with only nodes vectors (avoiding properties representation) and nodes_target with the type of each node
def delete_props(self):
self.nodes_pos = []
self.nodes_type = []
self.nodes_name = []
for t in self.n_types:
for n in self.n_types[t]:
if n in self.w2v:
self.nodes_pos.append(self.w2v[n])
self.nodes_type.append(t)
self.nodes_name.append(n)
print len(self.nodes_pos)
print len(self.nodes_type)
self.nodes_pos = list(self.nodes_pos)
self.nodes_type = list(self.nodes_type)
#Obtenemos el vector medio del traversal solicitado.
def get_vtraversal(self,traversal):
traversals = self.get_traversals(traversal,1)
total = 0
suma = "INICIO"
for t in traversals:
if t["t"] in self.w2v and t["s"] in self.w2v:
total += 1
vector = self.w2v[t["t"]] - self.w2v[t["s"]]
if suma == "INICIO":
suma = vector
else:
suma += vector
return suma / total
#Obtenemos una serie de traversals aleatorios del tipo indicado (tantos como indique 0<ts<1)
def get_traversals(self,traversal,ts):
if not os.path.exists("models/" + self.bd+"-trav-" + traversal + ".p"):
f = open( "models/" + self.bd+"-trav-" + traversal + ".p", "w" )
consulta = neo4j.CypherQuery(self.graph_db, "match (n)"+traversal+"(m) return n."+self.label+" as s,m."+self.label+" as t ,labels(m) as tipot").execute()
todas = []
for c in consulta:
todas.append({"s":c.s,"t":c.t,"tipot":c.tipot[0]})
pickle.dump(todas,f)
else:
f = open( "models/" + self.bd+"-trav-" + traversal + ".p", "r" )
todas = pickle.load(f)
for t in todas:
t["s"] = t["s"].replace(" ","_")
t["t"] = t["t"].replace(" ","_")
t["tipot"] = t["tipot"].replace(" ","_")
finales = random.sample(todas, int(len(todas)*ts))
return finales
def entity_retrieval(self,node,rel_type,target_t):
temp_pos = []
temp_name = []
linkstopredictV = []
self.r_analysis()
for idx,e in enumerate(self.nodes_type):
if e == target_t:
temp_pos.append(self.nodes_pos[idx])
temp_name.append(self.nodes_name[idx])
if len(temp_pos) < 1000:
ks = len(temp_pos)
else:
ks = 1000
clasificador = neighbors.KNeighborsClassifier(ks, "uniform",n_jobs=multiprocessing.cpu_count())
clasificador.fit(temp_pos, temp_name)
linkstopredictV.append(self.w2v[node]+self.m_vectors[str(rel_type)])
nbs = clasificador.kneighbors(linkstopredictV,ks,False)
result = []
for e in nbs[0]:
result.append(temp_name[e])
return result