def main():
source_file = codecs.open(sys.argv[1], encoding="utf-8", mode="r")
target_file = codecs.open(sys.argv[2], encoding="utf-8", mode="r")
source_out = codecs.open(sys.argv[3], encoding="utf-8", mode="w")
target_out = codecs.open(sys.argv[4], encoding="utf-8", mode="w")
#source = [line.lstrip() for line in source_file.readlines()]
#target = [line.lstrip() for line in target_file.readlines()]
source = source_file.readlines()
target = target_file.readlines()
aligner = PyGaleChurchAligner()
(cost, aligned_source, aligned_target) = aligner.align(source, target)
print cost, len(aligned_source), len(aligned_target)
for i in xrange(0, len(aligned_source)):
source_out.write(aligned_source[i] + "\n")
target_out.write(aligned_target[i] + "\n")
source_file.close()
target_file.close()
source_out.close()
target_out.close()
def main():
source_file = codecs.open(sys.argv[1], encoding="utf-8", mode="r")
target_file = codecs.open(sys.argv[2], encoding="utf-8", mode="r")
source_out = codecs.open(sys.argv[3], encoding="utf-8", mode="w")
target_out = codecs.open(sys.argv[4], encoding="utf-8", mode="w")
#source = [line.lstrip() for line in source_file.readlines()]
#target = [line.lstrip() for line in target_file.readlines()]
source = source_file.readlines()
target = target_file.readlines()
aligner = PyGaleChurchAligner()
(cost, aligned_source, aligned_target) = aligner.align(source, target)
print(cost, len(aligned_source), len(aligned_target))
for i in range(0, len(aligned_source)):
source_out.write(aligned_source[i] + "\n")
target_out.write(aligned_target[i] + "\n")
source_file.close()
target_file.close()
source_out.close()
target_out.close()
def main():
parser = optparse.OptionParser()
parser.add_option("-i", "--input-file", dest="input_file", default="",
type="string", help="Location of the uncompressed mined webpages")
parser.add_option("-n", "--num_entries", dest="num_entries", default=0,
type="int",
help="Maximum number of entries to examine, set to 0 for no limit")
parser.add_option("--language-pair", dest="language_pair",
default="English,Spanish",
help="Prints parallel data for the comma-separated language pair"
+ " when used with the --out-prefix option")
parser.add_option("--out-prefix", dest="out_prefix", default="",
help="Parallel data will be output to this location")
parser.add_option("--annotate", dest="annotate", default="",
help="Prints random pages to the given directory for annotation")
parser.add_option("--annotate_amount", dest="annotate_amount", default=100,
type="int", help="Number of document pairs to annotate")
parser.add_option("--annotation_file", dest="annotation_file", default="",
help="A file containing annotations of web page pairs")
parser.add_option("--annotation_dir", dest="annotation_dir", default="",
help="The location of the HTML for the annotated web pages")
(opts, args) = parser.parse_args()
# Initialize the HTML parsers
cleanup_parser = etree.HTMLParser(encoding = "utf-8",
target = parsers.CleanupTarget())
plaintext_parser = etree.HTMLParser(encoding = "utf-8",
target = parsers.PlaintextTarget())
strand_parser = etree.HTMLParser(encoding = "utf-8",
target = parsers.StrandTarget())
# Gale Church aligner
gc_aligner = PyGaleChurchAligner()
strand_aligner = strand.StrandAligner()
if opts.annotation_file and opts.annotation_dir:
data = read_annotated_data(strand_aligner, strand_parser,
opts.annotation_file, opts.annotation_dir)
folds = 5
features = []
true_pos = 0
false_pos = 0
total_pos = 0
correct = 0
total = len(data)
stats = {}
for fold in xrange(0, folds):
print "Fold %d:" % (fold + 1)
training_data = []
test_data = []
for i in xrange(0, len(data)):
if i % folds == fold:
test_data.append(data[i])
else:
training_data.append(data[i])
strand_aligner.me_model.set_training_data(training_data)
strand_aligner.me_model.lbfgs_train()
features.append(strand_aligner.me_model.get_features())
for example in test_data:
predicted_label = strand_aligner.me_model.get_label(example)
true_label = example.true_instance
print true_label, predicted_label
print strand_aligner.me_model.get_probs(example)
print example.instances[predicted_label]
if true_label == 0:
total_pos += 1
if predicted_label == 0:
true_pos += 1
elif true_label != predicted_label:
false_pos += 1
correct = true_pos + total - total_pos - false_pos
stats["Positives"] = total_pos
stats["Accuracy"] = (100.0 * correct) / total
stats["Precision"] = 0.0
if true_pos + false_pos > 0:
stats["Precision"] = (100.0 * true_pos) / (true_pos + false_pos)
stats["Recall"] = (100.0 * true_pos) / total_pos
stats["F1"] = 0.0
if stats["Recall"] + stats["Precision"] > 0.0:
stats["F1"] = 2 * stats["Precision"] * stats["Recall"] / (stats["Recall"]
+ stats["Precision"])
for i in xrange(0, len(features)):
print "Fold %d" % (i + 1)
print features[i]
print stats
if opts.input_file == "":
print "No input file given"
return
language_pair = None
if opts.language_pair:
languages = opts.language_pair.split(",")
if len(languages) != 2:
print "Error in language pair:", opts.language_pair
return
# TODO: Language codes
language_pair = (languages[0], languages[1])
data_to_annotate = []
aligned_strand_out = None
aligned_plaintext_out = None
plaintext_docs_out = None
segmenters = None
if opts.out_prefix and language_pair:
aligned_strand_out = []
aligned_plaintext_out = []
plaintext_docs_out = []
segmenters = []
for lang in language_pair:
aligned_strand_out.append(codecs.open(
"%s.strand.%s" % (opts.out_prefix, lang),
encoding="utf-8", mode="w"))
#aligned_plaintext_out.append(codecs.open(
# "%s.text.%s" % (opts.out_prefix, lang),
# encoding="utf-8", mode="w"))
#plaintext_docs_out.append(codecs.open(
# "%s.docs.%s" % (opts.out_prefix, lang),
# encoding="utf-8", mode="w"))
segmenters.append(Segmenter(lang))
in_file = open(opts.input_file, "r")
linecount = 0
for line in in_file:
(key, webpages) = parse_entry(line)
if len(key) == 0:
print "Malformed entry at line", linecount
else:
# default behavior for now: just print the URL
#print url_to_filename(key).encode('utf-8')
data_by_language = {}
for webpage in webpages:
if webpage['language'] not in data_by_language:
data_by_language[ webpage['language'] ] = {}
if opts.annotate:
try:
clean_html = apply_parser(webpage['html'], cleanup_parser)
data_by_language[ webpage['language'] ]["html"] = clean_html
except:
pass
if opts.out_prefix:
#plaintext = apply_parser(webpage['html'], plaintext_parser)
#data_by_language[ webpage['language'] ]["text"] = plaintext
lang = webpage['language']
if lang in language_pair:
try:
tagchunks = apply_parser(webpage['html'], strand_parser)
data_by_language[lang]["strand"] = tagchunks
except:
pass
if language_pair[0] in data_by_language and language_pair[1] in data_by_language:
if opts.annotate:
data_to_annotate.append((key,
data_by_language[ language_pair[0] ]["html"],
data_by_language[ language_pair[1] ]["html"]))
if opts.out_prefix and ("strand" in
data_by_language[ language_pair[0]]) and ("strand" in
data_by_language[ language_pair[1]]):
en_output = data_by_language[ language_pair[0] ]["strand"].split("\n")
es_output = data_by_language[ language_pair[1] ]["strand"].split("\n")
en_tagchunks = strand_aligner.create_tag_chunk_stream(en_output)
es_tagchunks = strand_aligner.create_tag_chunk_stream(es_output)
alignment = strand_aligner.align(en_tagchunks, es_tagchunks)
for (s, t) in alignment:
if (s and s.tc_type == strand.TCType.CHUNK
and t and t.tc_type == strand.TCType.CHUNK):
source_sents = segmenters[0].process(unicode(s.chunk_data))
target_sents = segmenters[1].process(unicode(t.chunk_data))
(cost, aligned_source, aligned_target) = gc_aligner.align(
source_sents, target_sents)
for i in xrange(0, len(aligned_source)):
s_sent = aligned_source[i]
t_sent = aligned_target[i]
if alpha_min_length(s_sent, t_sent) >= 5 and end_punc(s_sent, t_sent) == 1:
aligned_strand_out[0].write(s_sent + "\n")
aligned_strand_out[1].write(t_sent + "\n")
# Plain text output and alignment (TODO)
# Document output and alignment (TODO)
linecount += 1
if linecount == opts.num_entries:
break
in_file.close()
if opts.out_prefix:
for out_file in aligned_strand_out:
out_file.close()
for out_file in aligned_plaintext_out:
out_file.close()
for out_file in plaintext_docs_out:
out_file.close()
if opts.annotate:
mkdir_p(opts.annotate)
mkdir_p(opts.annotate + "/source")
mkdir_p(opts.annotate + "/target")
annotation_file = codecs.open(opts.annotate + "/annotation",
encoding="utf-8", mode="w")
shuffle(data_to_annotate)
for i in xrange(0, opts.annotate_amount):
(key, source, target) = data_to_annotate[i]
count_str = "%04d_" % i
out_source = codecs.open(opts.annotate + "/source/" + count_str +
url_to_filename(key), encoding="utf-8", mode="w")
out_target = codecs.open(opts.annotate + "/target/" + count_str +
url_to_filename(key), encoding="utf-8", mode="w")
out_source.write(source)
out_target.write(target)
annotation_file.write(key)
annotation_file.write("\n\n")
out_source.close()
out_target.close()
annotation_file.close()
