# parts = line[:-1].split('\t')
# label = int(parts[0])
# string = parts[1]
# augmented_sentences = eda.get_delete_sentences(string, n_aug=n_aug, alpha_rd=alpha)
# cosine_sim_list = utils_bert.compute_sent_similarities(string, augmented_sentences)
# dataset_cosine_sim_list += cosine_sim_list
# return dataset_cosine_sim_list
if __name__ == "__main__":
for method, name in [
#(compute_swap_similarity, 'swap'),
(compute_insert_similarity, 'insert'),
# (compute_insertswap_similarity, 'insertswap'),
]:
all_cosine_sim_list = []
for dataset_name in ['sst2', 'subj', 'trec']:
data_folder = config.data_folders[dataset_name]
delete_folder = utils_config.make_exp_folder(data_folder, 'delete')
delete_train_txt_path, _, _, _ = utils_config.get_txt_paths(
delete_folder)
dataset_cosine_sim_list = method(delete_train_txt_path)
all_cosine_sim_list += dataset_cosine_sim_list
print(
f"{name} has sim {mean(all_cosine_sim_list):.3f} stdev {stdev(all_cosine_sim_list):.3f}"
)
import config, utils_config, utils_mlp_a1_grad_sim_classification
from pathlib import Path
dataset_name = 'sst2'
data_folder = config.data_folders[dataset_name]
output_folder = Path("outputs")
exp_id = '9b'
num_classes = 2
resume_checkpoint_path = Path(
"checkpoints/sst2_vanilla_mlp_50_seednum0/e10_va0.7460.pt")
if __name__ == "__main__":
train_txt_path, train_embedding_path, test_txt_path, test_embedding_path = utils_config.get_txt_paths(
data_folder)
for train_subset in [500]: #, 20, 50]:
mean_val_acc, stdev_acc, mean_conf_acc, stdev_conf_acc, mlp_val_acc, mlp_stdev_acc, mlp_conf_acc, mlp_stdev_conf_acc = utils_mlp_a1_grad_sim_classification.train_mlp_multiple(
train_txt_path,
train_embedding_path,
test_txt_path,
test_embedding_path,
num_classes=num_classes,
dataset_name=dataset_name,
exp_id=exp_id,
train_subset=train_subset,
resume_checkpoint_path=resume_checkpoint_path,
num_seeds=1,
)
import config, utils_bert, utils_config, utils_svm, utils_processing
data_folder = config.data_folders['sst2']
if __name__ == "__main__":
train_txt_path, train_embedding_path, test_txt_path, test_embedding_path = utils_config.get_txt_paths(
data_folder)
# utils_bert.get_embedding_dict(train_txt_path, train_embedding_path)
# utils_bert.get_embedding_dict(test_txt_path, test_embedding_path)
insert_folder = utils_config.make_exp_folder(data_folder, f"insert-eval")
_, _, insert_test_txt_path, insert_test_embedding_path = utils_config.get_txt_paths(
insert_folder)
# utils_processing.augment_insert(test_txt_path, insert_test_txt_path, n_aug=2, alpha=0.3)
# utils_bert.get_embedding_dict(insert_test_txt_path, insert_test_embedding_path)
swap_folder = utils_config.make_exp_folder(data_folder, f"swap-eval")
_, _, swap_test_txt_path, swap_test_embedding_path = utils_config.get_txt_paths(
swap_folder)
# utils_processing.augment_swap(test_txt_path, swap_test_txt_path, n_aug=2, alpha=0.2)
# utils_bert.get_embedding_dict(swap_test_txt_path, swap_test_embedding_path)
# utils_svm.evaluate_svm_baselines( train_txt_path,
# test_txt_path,
# train_embedding_path,
# test_embedding_path,
# insert_test_txt_path,
# insert_test_embedding_path,
# swap_test_txt_path,
# swap_test_embedding_path,
