summaryrefslogtreecommitdiff
path: root/python/cdec/sa/extract.py
blob: 92d38af9537820ccd08532d950db65c35fd066d1 (plain)
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
#!/usr/bin/env python
import sys
import os
import re
import gzip
import argparse
import logging
import signal
import multiprocessing as mp
import cdec.sa
from cdec.sa._sa import monitor_cpu

extractor, prefix = None, None
online, compress = False, False

def make_extractor(args):
    global extractor, prefix, online, compress
    signal.signal(signal.SIGINT, signal.SIG_IGN) # Let parent process catch Ctrl+C
    load_features(args.features)
    extractor = cdec.sa.GrammarExtractor(args.config, online)
    prefix = args.grammars
    online = args.online
    compress = args.compress

def load_features(features):
    for featdef in features:
        logging.info('Loading additional feature definitions from %s', featdef)
        prefix = os.path.dirname(featdef)
        sys.path.append(prefix)
        __import__(os.path.basename(featdef).replace('.py', ''))
        sys.path.remove(prefix)

def extract(inp):
    global extractor, prefix, online, compress
    i, sentence = inp
    sentence = sentence[:-1]
    fields = re.split('\s*\|\|\|\s*', sentence)
    suffix = ''
    # 3 fields for online mode, 1 for normal
    if online:
        if len(fields) < 3:
            sys.stderr.write('Error: online mode requires references and alignments.'
                    '  Not adding sentence to training data: {}\n'.format(sentence))
            sentence = fields[0]
        else:
            sentence, reference, alignment = fields[0:3]
        if len(fields) > 3:
            suffix = ' ||| ' + ' ||| '.join(fields[3:])
    else:
        if len(fields) > 1:
            sentence = fields[0]
            suffix = ' ||| ' + ' ||| '.join(fields[1:])

    grammar_file = os.path.join(prefix, 'grammar.'+str(i))
    if compress: grammar_file += '.gz'
    with (gzip.open if compress else open)(grammar_file, 'w') as output:
        for rule in extractor.grammar(sentence):
            output.write(str(rule)+'\n')
    # Add training instance _after_ extracting grammars
    if online:
        extractor.add_instance(sentence, reference, alignment)
    grammar_file = os.path.abspath(grammar_file)
    return '<seg grammar="{}" id="{}">{}</seg>{}'.format(grammar_file, i, sentence, suffix)

def stream_extract():
    global extractor, online, compress
    while True:
        line = sys.stdin.readline()
        if not line:
            break
        fields = re.split('\s*\|\|\|\s*', line.strip())
        # context ||| cmd
        if len(fields) == 2:
            (context, cmd) = fields
            if cmd.lower() == 'drop':
                if online:
                    extractor.drop_ctx(context)
                else:
                    sys.stderr.write('Error: online mode not set. Skipping line: {}\n'.format(line.strip()))
        # context ||| sentence ||| grammar_file
        elif len(fields) == 3:
            (context, sentence, grammar_file) = fields
            with (gzip.open if compress else open)(grammar_file, 'w') as output:
                for rule in extractor.grammar(sentence, context):
                    output.write(str(rule)+'\n')
        # context ||| sentence ||| reference ||| alignment
        elif len(fields) == 4:
            (context, sentence, reference, alignment) = fields
            if online:
                extractor.add_instance(sentence, reference, alignment, context)
            else:
                sys.stderr.write('Error: online mode not set. Skipping line: {}\n'.format(line.strip()))
        else:
            sys.stderr.write('Error: see README.md for stream mode usage.  Skipping line: {}\n'.format(line.strip()))

def main():
    global online
    logging.basicConfig(level=logging.INFO)
    parser = argparse.ArgumentParser(description='Extract grammars from a compiled corpus.')
    parser.add_argument('-c', '--config', required=True,
                        help='extractor configuration')
    parser.add_argument('-g', '--grammars',
                        help='grammar output path')
    parser.add_argument('-j', '--jobs', type=int, default=1,
                        help='number of parallel extractors')
    parser.add_argument('-s', '--chunksize', type=int, default=10,
                        help='number of sentences / chunk')
    parser.add_argument('-f', '--features', nargs='*', default=[],
                        help='additional feature definitions')
    parser.add_argument('-o', '--online', action='store_true',
                        help='online grammar extraction')
    parser.add_argument('-z', '--compress', action='store_true',
                        help='compress grammars with gzip')
    parser.add_argument('-t', '--stream', action='store_true',
                        help='stream mode (see README.md)')
    args = parser.parse_args()

    if not (args.grammars or args.stream):
        sys.stderr.write('Error: either -g/--grammars or -t/--stream required\n')
        sys.exit(1)

    if args.grammars and not os.path.exists(args.grammars):
        os.mkdir(args.grammars)
    for featdef in args.features:
        if not featdef.endswith('.py'):
            sys.stderr.write('Error: feature definition file <{}>'
                    ' should be a python module\n'.format(featdef))
            sys.exit(1)

    online = args.online
    stream = args.stream

    start_time = monitor_cpu()
    if args.jobs > 1:
        if stream:
            sys.stderr.write('Error: stream mode incompatible with multiple jobs\n')
            sys.exit(1)
        logging.info('Starting %d workers; chunk size: %d', args.jobs, args.chunksize)
        pool = mp.Pool(args.jobs, make_extractor, (args,))
        try:
            for output in pool.imap(extract, enumerate(sys.stdin), args.chunksize):
                print(output)
        except KeyboardInterrupt:
            pool.terminate()
    else:
        make_extractor(args)
        if stream:
            stream_extract()
        else:
            for output in map(extract, enumerate(sys.stdin)):
                print(output)

    stop_time = monitor_cpu()
    logging.info("Overall extraction step took %f seconds", stop_time - start_time)

if __name__ == '__main__':
    main()