def copy_config(self):
'''
Copies the config to use for different grid search configurations
'''
c = Config()
c.__dict__ = self.__dict__.copy()
return c
torch.nn.utils.clip_grad_norm_(model.parameters(), config.clip)
optimizer.step()
# Save model checkpoint
if train_num_batches > 0 and train_num_batches % config.eval_every_minibatch == 0:
model.eval()
dev_evaluator.evaluate(model, train_num_batches, float(loss))
if train_num_batches > config.num_batches:
break
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("-c", "--config_file", required=True)
parser.add_argument("-g", "--grid_search", default="False")
args = parser.parse_args()
# Set up the config
config = Config(args.config_file)
# For non grid search, must set up exp dir
if args.grid_search == "False":
exp_dir = make_exp_dir(config)
copytree(os.path.join(os.environ['SED_ROOT'], 'src'), os.path.join(exp_dir, 'src'))
config.save_config(exp_dir)
else:
exp_dir = os.path.split(args.config_file)[0]
train_model(config, exp_dir)
value = sdtw.compute()
dtw_mat[i] = torch.from_numpy(sdtw.grad())
return dtw_mat
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("-t", "--test_file", required=True)
parser.add_argument("-s", "--stance_exp_dir", required=True)
parser.add_argument("-d", "--dtw_exp_dir", required=True)
parser.add_argument("-o", "--output_folder", required = True)
args = parser.parse_args()
output_folder = args.output_folder
config = Config(os.path.join(args.stance_exp_dir, "config.json"))
tokenizer, max_len_token = get_tokenizer(config)
vocab = get_vocab(config, tokenizer, max_len_token)
qry_tk, cnd_tk, qry_len, cnd_len = load_data(args.test_file, vocab, config, tokenizer, max_len_token)
stance_sim, stance_dist, stance_pi, stance_mulpld = print_stance_mm(args.stance_exp_dir, qry_tk, cnd_tk)
dtw_mat = print_dtw_mm(args.dtw_exp_dir, qry_tk, cnd_tk)
stance_sim, stance_dist, stance_pi, stance_mulpld, dtw_mat = stance_sim.cpu().data.numpy(), stance_dist.cpu().data.numpy(),\
stance_pi.cpu().data.numpy(), stance_mulpld.cpu().data.numpy(), \
dtw_mat.cpu().data.numpy()
min_sim = np.min(stance_sim)
max_sim = np.max(stance_sim)
min_pi = np.min(stance_pi)
max_pi = np.max(stance_pi)
min_dist = np.min(stance_dist)
labeled_file=test_label_file, output_file=test_output_file)
model = torch.load(os.path.join(exp_dir, "best_model"))
model.eval()
test_evaluator.evaluate(model, train_num_batches=-1, train_loss=-1)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("-c", "--config_dir", required=True)
parser.add_argument("-p", "--is_parallel", required=True)
parser.add_argument("-s", "--shard")
args = parser.parse_args()
config_file = os.path.join(args.config_dir, "config.json")
config = Config(config_file)
if args.is_parallel == "True":
data_dir = os.path.split(config.test_file)[0]
test_output_dir = os.path.join(args.config_dir, "test_shards")
if not os.path.exists(test_output_dir):
os.makedirs(test_output_dir)
test_output_file = os.path.join(test_output_dir,
"shard_" + args.shard + ".pred")
test_label_file = os.path.join(data_dir, "test_shards",
"shard_" + args.shard)
else:
test_output_file = os.path.join(args.config_dir, "test.pred")
test_label_file = config.test_file
optimizer.step()
if train_num_batches == config.num_minibatches:
break
if train_num_batches > 0 and train_num_batches % config.eval_every_minibatch == 0:
model.eval()
dev_evaluator.evaluate(model, train_num_batches)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("-c", "--config_file", required=True)
parser.add_argument("-g", "--grid_search", default="False")
args = parser.parse_args()
# Set up the config
config = Config(args.config_file)
config.update_dataset()
# For non grid search, must set up exp dir
if args.grid_search == "False":
exp_dir = make_exp_dir(config.dataset_name, config.model_name,
config.tokenizer_name)
copytree(os.path.join(os.environ['SED_ROOT'], 'src'),
os.path.join(exp_dir, 'src'))
config.save_config(exp_dir)
else:
exp_dir = os.path.split(args.config_file)[0]
train_model(config, exp_dir)
from main.objects.DataExtractor import DataExtractor
from main.objects.Tokenizer import Tokenizer
from main.utils.model_helper import get_baseline
from main.utils.util import make_exp_dir
def test_model(config, exp_dir, test_label_file=None, test_output_file=None):
test_evaluator = Evaluator(config, 'test', exp_dir, list_k=[1, 10, 50, 100], \
labeled_file=test_label_file, output_file=test_output_file)
data_extractor = DataExtractor(config)
tokenizer = Tokenizer(config)
model = get_baseline(config, data_extractor, tokenizer)
test_evaluator.evaluate(model, train_num_batches=-1, train_loss=-1)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("-c", "--config_file", required=True)
parser.add_argument("-f", "--fold")
args = parser.parse_args()
config = Config(args.config_file)
exp_dir = os.path.dirname(args.config_file)
test_output_file = os.path.join(exp_dir, "test_fold_%d.pred" % int(args.fold))
test_label_file = os.path.join("data", "cross_validation", "fold_%d" % int(args.fold), "test.data")
test_model(config, exp_dir, test_label_file, test_output_file)