data = read_twitter(trainfile=args.train, evalfiles=args.eval)
import tagger
if args.tagger == "logreg":
tagger = tagger.LogisticRegressionTagger()
elif args.tagger == "crf":
tagger = tagger.CRFPerceptron()
else:
sys.stderr.write("Did not properly select tagger!")
sys.exit(1)
# Train the tagger
tagger.fit_data(data.train_sents, data.train_labels)
# Evaluation (also writes out predictions)
trainoutfile = "{}/{}.pred".format(args.outdir,
os.path.basename(args.train))
print "### Train evaluation; writing to {}".format(trainoutfile)
data.train_preds = tagger.evaluate_data(data.train_sents,
data.train_labels)
write_preds(trainoutfile, data.train_sents, data.train_labels,
data.train_preds)
for evalstr, evalset in zip(args.eval, data.eval):
evaloutfile = "{}/{}.pred".format(args.outdir,
os.path.basename(evalstr))
print "### evaluation of {}; writing to {}".format(
evalstr, evaloutfile)
preds = tagger.evaluate_data(evalset["sents"], evalset["labels"])
write_preds(evaloutfile, evalset["sents"], evalset["labels"], preds)
use_test = args.test
data = read_twitter(test=use_test)
if model == 'crf':
tagger = tagger.CRFPerceptron()
else:
tagger = tagger.LogisticRegressionTagger()
# Train the tagger
tagger.fit_data(data.train_sents, data.train_labels)
# Evaluation (also writes out predictions)
print "### Train evaluation"
data.train_preds = tagger.evaluate_data(data.train_sents, data.train_labels)
write_preds("%s/twitter_train.%s.pred" % (base_path_predictions, model),
data.train_sents,
data.train_labels,
data.train_preds)
print "### Dev evaluation"
data.dev_preds = tagger.evaluate_data(data.dev_sents, data.dev_labels)
write_preds("%s/twitter_dev.%s.pred" % (base_path_predictions, model),
data.dev_sents, data.dev_labels, data.dev_preds)
# Following is commented, only useful once test data is available.
if use_test:
print "### Generating Test predictions"
data.test_preds = tagger.evaluate_data(data.test_sents, data.test_labels, quite=True)
write_preds("%s/twitter_test.%s.pred" % (base_path_predictions, model),
# Do no run, the following function was used to generate the splits
# file_splitter("data/twitter_train_all.pos", "data/twitter_train.pos", "data/twitter_dev.pos")
dname = "pos" # or "ner"
data = read_twitter(dname)
# data = synthetic_data()
import tagger
tagger = tagger.LogisticRegressionTagger()
#tagger = tagger.CRFPerceptron()
# Train the tagger
tagger.fit_data(data.train_sents, data.train_labels)
# Evaluation (also writes out predictions)
print "### Train evaluation"
data.train_preds = tagger.evaluate_data(data.train_sents,
data.train_labels)
write_preds("data/twitter_train.%s.pred" % dname, data.train_sents,
data.train_labels, data.train_preds)
print "### Dev evaluation"
data.dev_preds = tagger.evaluate_data(data.dev_sents, data.dev_labels)
write_preds("data/twitter_dev.%s.pred" % dname, data.dev_sents,
data.dev_labels, data.dev_preds)
# Following is commented, only useful once test data is available.
# print "### Test evaluation"
# data.test_preds = tagger.evaluate_data(data.test_sents, data.test_labels)
# write_preds("data/twitter_test.%s.pred" % dname,
# data.test_sents, data.test_labels, data.test_preds)
default=4,
help="batch size")
try:
args = parser.parse_args()
except IOError as msg:
parser.error(str(msg))
# print os.getcwd() #
data = read_twitter(trainfile=args.train, evalfiles=args.eval)
import tagger
if args.tagger == "logreg":
tagger = tagger.LogisticRegressionTagger()
elif args.tagger == "crf":
tagger = tagger.CRFPerceptron(args.epochs, args.batch_size)
else:
sys.stderr.write("Did not properly select tagger!")
sys.exit(1)
# Train the tagger
tagger.fit_data(data.train_sents, data.train_labels)
# Evaluation (also writes out predictions)
for evalstr, evalset in zip(args.eval, data.eval):
evaloutfile = "{}/{}.pred".format(args.outdir,
os.path.basename(evalstr))
print("### evaluation of {}; writing to {}".format(
evalstr, evaloutfile))
preds = tagger.evaluate_data(evalset["sents"], evalset["labels"])
write_preds(evaloutfile, evalset["sents"], evalset["labels"], preds)
