def main():
"""
Executes the baseline model. This loads the training data, training labels, and dev data, then trains a logistic
regression model, then dumps predictions to the specified file.
Modify the middle of this code, between the two commented blocks, to create your own model.
"""
parser = argparse.ArgumentParser(
description='Duolingo shared task baseline model')
parser.add_argument('--language',
default='en_es',
help='choose from [es_en, en_es, fr_en]',
required=False)
parser.add_argument('--dataset_path',
default='../data/%s/',
required=False)
parser.add_argument('--outputs_path', default='./outputs/', required=False)
args = parser.parse_args()
dataset_path = args.dataset_path % args.language
assert os.path.isdir(dataset_path)
train_path = dataset_path + '%s.slam.20190204.train' % args.language
dev_path = dataset_path + '%s.slam.20190204.dev' % args.language
test_path = dataset_path + '%s.slam.20190204.test' % args.language
assert os.path.isfile(train_path)
assert os.path.isfile(dev_path)
assert os.path.isfile(test_path)
if not os.path.isdir(args.outputs_path): os.mkdir(args.outputs_path)
# ============================== Hyper Parameter ==============================
dbg = False
from_path = None
# from_path = './saved_model/seq2seq_nomlp_20'
# from_path = './saved_model/seq2seq_exp_20'
# from_path = './saved_model/attention_v2_20'
# from_path = './saved_model/cnn_3'
# from_path = './saved_model/seq2seq_c_7'
# from_path = './saved_model/transformer_1'
epochs = 10 if dbg else 10
lang = Lang()
# ============================== Data Loading ==============================
print('Begin Data Loading')
start_time = time.time()
training_data, training_labels = load_data(train_path,
lang,
dbg=dbg,
use_all_features=True)
dev_data = load_data(dev_path, lang, use_all_features=True)
test_data = load_data(test_path, lang, use_all_features=True)
users = list(
get_users(training_data).union(get_users(dev_data)).union(
get_users(test_data)))
lang.addUsers(users)
for i in range(len(lang.letters)):
lang.letter2Index[lang.letters[i]] = i
end_time = time.time()
print('Data Loaded\t Time Taken %0.2fm' % ((end_time - start_time) / 60))
model = Model(lang)
# ============================== Training ==============================
if from_path == None:
print('Begin Training')
train_loader = get_dataloader(training_data, lang, training_labels)
model.train(train_loader, epochs)
# ============================== Inference ==============================
# print('Begin Inference-Dev', end=' ')
# start_time = time.time()
# dev_loader = get_dataloader(dev_data, lang)
# predictions = model.predict_for_set(dev_loader, from_path)
# with open(args.outputs_path + '%s_dev_predictions.pred' % args.language, 'wt') as f:
# for instance_id, prediction in iteritems(predictions):
# f.write(instance_id + ' ' + str(prediction) + '\n')
# end_time = time.time()
# print('| %0.2fm' % ((end_time-start_time)/60))
print('Begin Inference-Test', end=' ')
start_time = time.time()
test_loader = get_dataloader(test_data, lang)
predictions = model.predict_for_set(test_loader, from_path)
with open(args.outputs_path + '%s_test_predictions.pred' % args.language,
'wt') as f:
for instance_id, prediction in predictions.items():
f.write(instance_id + ' ' + str(prediction) + '\n')
end_time = time.time()
print('| %0.2fm' % ((end_time - start_time) / 60))
