def sub_run(sub_train_path, sub_model_path, raw_test_path, sub_tag_path, test_corpus, sub_corpus):
# Next, train the parser.
stanford_tagger.train(sub_train_path, sub_model_path)
# Now, run it.
stanford_tagger.test(raw_test_path, sub_model_path, sub_tag_path)
num_tokens = sum([len(i) for i in sub_corpus])
# Load the result of the tagging...
result_corpus = POSCorpus.read_slashtags(sub_tag_path)
acc = pos_eval.poseval(result_corpus, test_corpus)
return (num_tokens, acc)
def full_run(c):
# The step of sentences by which to increment.
step_increment = 50
curve_dir = os.path.abspath(writedir(c["curve_dir"]))
train_path = c["train_path"]
test_path = c["test_path"]
train_corpus = POSCorpus.read_slashtags(train_path)
test_corpus = POSCorpus.read_slashtags(test_path)
# Let's go ahead and strip the tags from the test corpus.
raw_test_name = "test_data.txt"
raw_test_path = os.path.join(curve_dir, raw_test_name)
test_corpus.write(raw_test_name, "raw", outdir=curve_dir)
# Now, let's add 100 sentences at a time until we max out.
sent_limit = 0
p = Pool(8)
results = {}
while sent_limit < len(train_corpus):
# Adding 100 to the limit, starting from zero, means
# we will get the last <99 instances too.
actual_limit = sent_limit + step_increment
sub_corpus = POSCorpus(train_corpus[0:actual_limit])
# Let's make the x values the number of tokens instead of sentences...
# Let's create the necessary filenames.
sub_train_path = os.path.join(curve_dir, "%d_train.txt" % actual_limit)
sub_model_path = os.path.join(curve_dir, "%d_train.model" % actual_limit)
sub_tag_path = os.path.join(curve_dir, "%d_tagged.txt" % actual_limit)
# Get the number of tokens in the corpus for our x axis...
num_tokens = sum([len(x) for x in sub_corpus])
sub_corpus.write(os.path.basename(sub_train_path), "slashtags", outdir=curve_dir)
p.apply_async(
sub_run,
args=[sub_train_path, sub_model_path, raw_test_path, sub_tag_path, test_corpus, sub_corpus],
callback=lambda x: results.update({x[0]: x[1]}),
)
# Now, increase the sentence limit
sent_limit += step_increment
p.close()
p.join()
# Also, define where we will store the curve points.
curve_points = "curve_data.txt"
curve_f = open(os.path.join(curve_dir, curve_points), "w")
for size, acc in sorted(results.items(), key=lambda x: x[0]):
curve_f.write("%d,%.2f\n" % (size, acc))
curve_f.close()
