filter_abstracts_binClass.py
5.6 KB
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
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.decomposition import TruncatedSVD
from sklearn.model_selection import GridSearchCV
from sklearn import metrics
from sklearn.svm import SVC
import numpy as np
import argparse
import csv
import os
from sklearn.externals import joblib
from time import time
def get_abstracts(file_name, label):
f = open(file_name)
extract = {}
docs = []
empties = []
lines = f.readlines()
copyright = False
for i, ln in enumerate(lines):
if not ln.strip():
empties.append(i)
continue
elif ' doi: ' in ln:
for j in range(i, i + 10):
if not lines[j].strip():
title_idx = j + 1
break
continue
elif 'Copyright ' in ln or 'Publish' in ln or u'\N{COPYRIGHT SIGN}' in ln:
copyright = True
elif 'DOI: ' in ln:
if 'PMCID: ' in lines[i + 1]:
extract['pmid'] = int(lines[i + 2].strip().split()[1])
elif not 'PMCID: ' in lines[i + 1] and 'PMID: ' in lines[i + 1]:
extract['pmid'] = int(lines[i + 1].strip().split()[1])
if copyright:
get = slice(empties[-3], empties[-2])
copyright = False
else:
get = slice(empties[-2], empties[-1])
extract['body'] = " ".join(lines[get]).replace("\n", ' '
).replace(" ", ' ')
title = []
for j in range(title_idx, title_idx + 5):
if lines[j].strip():
title.append(lines[j])
else:
break
extract['title'] = " ".join(title).replace("\n", ' '
).replace(" ", ' ')
extract['topic'] = label
docs.append(extract)
empties = []
extract = {}
return docs
parser = argparse.ArgumentParser(
description="This script separates abstracts of biomedical papers that"
"report data from biomedical experiments from those that do not.")
parser.add_argument("--input", help="Input file containing the abstracts to"
"be predited.")
parser.add_argument("--classA", help="Input file containing the abstracts of"
" class useless to be learned.")
parser.add_argument("--classB", help="Input file containing the abstracts of"
" class USEFUL to be learned.")
parser.add_argument("--out", help="Path to the output directory "
"(default='./filter_output')", default="filter_output")
parser.add_argument("--svcmodel", help="Path to custom pretrained svc model"
"(default='./model_binClass/svm_model.pkl')",
default="model_binClass/svm_model.pkl")
args = parser.parse_args()
labels = {0: 'useless', 1: 'useful'}
vectorizer = TfidfVectorizer(binary=True)
if args.classA and args.classB and not args.input:
f0 = open("model_params_binClass.conf")
n_iter_search = 10
params = [p for p in csv.DictReader(f0)]
f0.close()
names = list(params[0].keys())
model_params = {n: [] for n in names}
for n in names:
for d in params:
for k in d:
if k == n:
try:
model_params[n].append(float(d[k]))
except ValueError:
model_params[n].append(d[k])
model_params = {k: list(set(model_params[k])) for k in model_params}
abstracs = get_abstracts(file_name=args.classA, label=labels[0])
abstracs += get_abstracts(file_name=args.classB, label=labels[1])
tfidf_model = vectorizer.fit([x['body'] for x in abstracs])
X = tfidf_model.transform([x['body'] for x in abstracs])
svd = TruncatedSVD(n_components=200, random_state=42, n_iter=20)
svd_model = svd.fit(X)
X = svd_model.transform(X)
y = [0 if x['topic'] == 'useless' else 1 for x in abstracs]
clf = SVC()
clf = GridSearchCV(clf, cv=3, param_grid=model_params,
n_jobs=-1, scoring='f1')
start = time()
clf.fit(X, y)
print("GridSearch took %.2f seconds for %d candidates"
" parameter settings." % ((time() - start), n_iter_search))
print()
print("The best model parameters:")
print(vectorizer)
print(svd)
print(clf.best_estimator_)
print()
print("The best F1 score:")
print(clf.best_score_)
joblib.dump(clf.best_estimator_, 'model_binClass/svm_model.pkl')
joblib.dump(tfidf_model, 'model_binClass/tfidf_model.pkl')
joblib.dump(svd_model, 'model_binClass/svd_model.pkl')
else:
clf = joblib.load(args.svcmodel)
vectorizer = joblib.load('model_binClass/tfidf_model.pkl')
svd = joblib.load('model_binClass/svd_model.pkl')
print(vectorizer)
print(svd)
print(clf)
abstracs = get_abstracts(file_name=args.input, label='unknown')
X = vectorizer.transform([x['body'] for x in abstracs])
X = svd.transform(X)
classes = clf.predict(X)
if not os.path.exists(args.out):
os.makedirs(args.out)
# Writing predictions to output files
with open(args.out + "/" + labels[0] + ".out", 'w') as f0, \
open(args.out + "/" + labels[1] + ".out", 'w') as f1:
for c, a in zip(classes, abstracs):
if c == 0:
f0.write("%d\t%s\n" % (a['pmid'], a['body']))
elif c == 1:
f1.write("%d\t%s\n" % (a['pmid'], a['body']))
print ("FINISHED!!")