/
create_model.py
163 lines (127 loc) · 4.48 KB
/
create_model.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
import numpy as np
from multiprocessing import Pool
import networkx as nx
import itertools
from helpers import parallel_execute_query, create_CPD, index_sort
import models
import cPickle as pickle
import sys
def eval_likelihood(self, Y, debug=False):
return models.TreeModel.eval_likelihood(self, Y, do_check=False)
if hasattr(self, 'NONPARAM'):
Y = tuple([i for i in xrange(len(Y)) if Y[i] == 1])
return np.log(self.Y_counts[Y])
lprob = 0
for i,l in enumerate(self.latents):
parent_states = []
parents = self.parents_of[l]
for p in parents:
parent_index = self.latent_lookup[p]
parent_states.append(Y[parent_index])
key = tuple([Y[i]]+parent_states)
#if debug:
# print l, self.parents_of[l], key, self.lCPD[l][key]
lprob += self.lCPD[l][key]
assert lprob == models.TreeModel.eval_likelihood(self, Y, do_check=False)
return lprob
networkdir = sys.argv[1].strip('/')
print 'networkdir', networkdir
N = int(networkdir.split('-')[-2].strip('N'))
args = networkdir.split('-')[1:]
if 'local_anchors' in networkdir:
anchor_source = 'local'
else:
anchor_source = 'simple'
if 'indep' in networkdir:
complexity = 1
elif 'third_order' in networkdir:
complexity = 3
else:
complexity = 2
header = pickle.load(file(networkdir+'/pickles/header.pk'))
inv_header = dict(zip(header, xrange(len(header))))
latents = pickle.load(file(networkdir+'/pickles/labels.pk'))
labels = latents
latents = [header[l] for l in latents]
L = len(labels)
adj = np.loadtxt(networkdir+'/'+'.'.join(args)+'.'+anchor_source+'_anchors.'+str(N)+'.bn_mat')
print adj.sum(0)
best_score = 0
root = [i for i,v in enumerate(adj.sum(0)) if v==0]
print 'root is', root
parents_of = {}
children_of = {}
queries = []
for j in xrange(L):
parents_of[latents[j]] = [latents[i] for i in np.nonzero(adj[:,j])[0]]
children_of[latents[j]] = [latents[i] for i in np.nonzero(adj[j,:])[0]]
queries.append(tuple([inv_header[latents[j]]] + [inv_header[i] for i in parents_of[latents[j]]]))
moments = pickle.load(file(networkdir+'/pickles/estimated_moments.'+'.'.join(args)+'.'+anchor_source+'_anchors'+str(N)+'.'+str(complexity)+'.pk'))
counter = {}
for q in queries:
counter[q] = moments[tuple(sorted(q))].transpose(index_sort(q))
CPDs = {}
lCPDs = {}
for k,val in counter.items():
CPDs[header[k[0]], tuple([header[z] for z in k[1:]])] = create_CPD(np.array(val + 1e-6, dtype=float))
model = models.TreeModel(CPDs, latents, format='CPDs')
for k in sorted(model.lCPD):
print k, model.lCPD[k]
model_ll = 0
data = file(networkdir+'/samples/'+str(N)+'_samples.dat').readlines()
def eval(dat):
if len(dat) == 0:
return 0
dat = dat.split()[2:]
Y = [int(str(z) in dat) for z in labels]
e = eval_likelihood(model, Y, debug=True)
return e
pool = Pool(48)
for i,e in enumerate(pool.imap(eval, data[:N])):
if i % 10000 == 0:
print i
model_ll += e
#for d,dat in enumerate(data[:N]):
# if d % 10000 == 0:
# print d
# if len(dat) == 0:
# continue
# try:
# assert dat.split()[1] == 'compact'
# except:
# print 'could not read', dat
# continue
# dat = dat.split()[2:]
# Y = [int(str(z) in dat) for z in labels]
# e = eval_likelihood(model, Y, debug=True)
print 'train', model_ll / N
model_ll = 0
for dat in data[-5000:]:
assert dat.split()[1] == 'compact'
dat = set([int(z) for z in dat.strip().split()[2:]])
Y = [int(z in dat) for z in labels]
e = eval_likelihood(model, Y, debug=True)
model_ll += e
print 'heldout', model_ll / 5000.0
print 'saving adj matrix'
np.savetxt(networkdir+'/results/adjacency.mat', adj)
pickle.dump(model, file(networkdir+'/results/skeleton.pk', 'w'))
moments = parallel_execute_query([tuple(sorted(q)) for q in queries], networkdir, N, len(header), 10)
for k,val in moments.items():
moments[k] = val / float(val.sum())
counter = {}
for q in queries:
counter[q] = moments[tuple(sorted(q))].transpose(index_sort(q))
CPDs = {}
lCPDs = {}
for k,val in counter.items():
CPDs[header[k[0]], tuple([header[z] for z in k[1:]])] = create_CPD(np.array(val + 1e-6, dtype=float))
model = models.TreeModel(CPDs, latents, format='CPDs')
model_ll = 0
for dat in data[-5000:]:
assert dat.split()[1] == 'compact'
dat = set([int(z) for z in dat.strip().split()[2:]])
Y = [int(z in dat) for z in labels]
e = eval_likelihood(model, Y, debug=True)
model_ll += e
print 'optimistic heldout', model_ll / 5000.0