Spaces:
Running
Running
File size: 35,744 Bytes
6ebf426 2d06d0e 6ebf426 144e87a 3bdbad7 6ebf426 ac8e861 6ebf426 2d06d0e 6ebf426 2d06d0e 6ebf426 5b98825 6ebf426 5b98825 6ebf426 2d06d0e 6ebf426 3bdbad7 6ebf426 3bdbad7 144e87a ffb4b87 144e87a ffb4b87 144e87a be26700 144e87a be26700 144e87a 6ebf426 0ad4d45 6ebf426 3bdbad7 6ebf426 3bdbad7 6ebf426 144e87a 6ebf426 144e87a 6ebf426 144e87a 6ebf426 144e87a 6ebf426 144e87a 6ebf426 144e87a f4725ed 693e74a 6ebf426 5799ee8 6ebf426 3bdbad7 6ebf426 3bdbad7 6ebf426 3bdbad7 6ebf426 3bdbad7 6ebf426 144e87a ffb4b87 144e87a ffb4b87 144e87a ffb4b87 144e87a ffb4b87 144e87a 6ebf426 3bdbad7 6ebf426 988da9d 144e87a be26700 144e87a be26700 144e87a 9a26761 3bdbad7 144e87a 55fc739 3bdbad7 6ebf426 be26700 6ebf426 be26700 6ebf426 3bdbad7 144e87a 6ebf426 adbaefa |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 |
#%%
import gradio as gr
import time
import sys
import os
import torch
import torch.backends.cudnn as cudnn
import numpy as np
import json
import networkx as nx
import spacy
# os.system("python -m spacy download en-core-web-sm")
import pickle as pkl
from tqdm import tqdm
import traceback
#%%
# please use torch.load with map_location=torch.device('cpu') to map your storages to the CPU.
# torch.loa
from torch.nn.modules.loss import CrossEntropyLoss
from transformers import AutoTokenizer
from transformers import BioGptForCausalLM, BartForConditionalGeneration
from server import server_utils
import Parameters
from Openai.chat import generate_abstract
from DiseaseSpecific import utils, attack
from DiseaseSpecific.attack import calculate_edge_bound, get_model_loss_without_softmax
specific_model = None
def capitalize_the_first_letter(s):
return s[0].upper() + s[1:]
parser = utils.get_argument_parser()
parser = utils.add_attack_parameters(parser)
parser.add_argument('--init-mode', type = str, default='single', help = 'How to select target nodes') # 'single' for case study
args = parser.parse_args()
args = utils.set_hyperparams(args)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# device = torch.device("cpu")
args.device = device
args.device1 = device
if torch.cuda.device_count() >= 2:
args.device = "cuda:0"
args.device1 = "cuda:1"
utils.seed_all(args.seed)
np.set_printoptions(precision=5)
cudnn.benchmark = False
model_name = '{0}_{1}_{2}_{3}_{4}'.format(args.model, args.embedding_dim, args.input_drop, args.hidden_drop, args.feat_drop)
model_path = 'DiseaseSpecific/saved_models/{0}_{1}.model'.format(args.data, model_name)
data_path = os.path.join('DiseaseSpecific/processed_data', args.data)
data = utils.load_data(os.path.join(data_path, 'all.txt'))
print('done')
n_ent, n_rel, ent_to_id, rel_to_id = utils.generate_dicts(data_path)
with open(os.path.join(data_path, 'filter.pickle'), 'rb') as fl:
filters = pkl.load(fl)
with open(os.path.join(data_path, 'entityid_to_nodetype.json'), 'r') as fl:
entityid_to_nodetype = json.load(fl)
with open(os.path.join(data_path, 'edge_nghbrs.pickle'), 'rb') as fl:
edge_nghbrs = pkl.load(fl)
with open(os.path.join(data_path, 'disease_meshid.pickle'), 'rb') as fl:
disease_meshid = pkl.load(fl)
with open(os.path.join(data_path, 'entities_dict.json'), 'r') as fl:
entity_to_id = json.load(fl)
with open(Parameters.GNBRfile+'entity_raw_name', 'rb') as fl:
entity_raw_name = pkl.load(fl)
with open(os.path.join(data_path, 'entities_reverse_dict.json'), 'r') as fl:
id_to_entity = json.load(fl)
id_to_meshid = id_to_entity.copy()
with open(Parameters.GNBRfile+'retieve_sentence_through_edgetype', 'rb') as fl:
retieve_sentence_through_edgetype = pkl.load(fl)
with open(Parameters.GNBRfile+'raw_text_of_each_sentence', 'rb') as fl:
raw_text_sen = pkl.load(fl)
with open(Parameters.UMLSfile+'drug_term', 'rb') as fl:
drug_term = pkl.load(fl)
print('done')
gallery_specific_target_path = os.path.join(data_path, 'DD_target_distmult_GNBR_random_50_exists:False_single.txt')
gallery_specific_link_path = 'DiseaseSpecific/attack_results/GNBR/cos_distmult_random_50_exists:False_20_quadratic_single_0.5.txt'
gallery_specific_text_path = 'DiseaseSpecific/generate_abstract/random_0.5_bioBART_finetune.json'
gallery_agnostic_target_path = 'DiseaseAgnostic/processed_data/target_0.7random.pkl'
gallery_agnostic_link_path = 'DiseaseAgnostic/processed_data/attack_edge_distmult_0.7random.pkl'
gallery_agnostic_text_path = 'DiseaseAgnostic/generate_abstract/random0.7_bioBART_finetune.json'
gallery_specific_chat_path = 'DiseaseSpecific/generate_abstract/random_0.5_chat.json'
gallery_agnostic_chat_path = 'DiseaseAgnostic/generate_abstract/random0.7_chat.json'
gallery_specific_target = utils.load_data(gallery_specific_target_path, drop=False)
gallery_specific_link = utils.load_data(gallery_specific_link_path, drop=False)
with open(gallery_specific_text_path, 'r') as fl:
gallery_specific_text = json.load(fl)
with open(gallery_agnostic_target_path, 'rb') as fl:
gallery_agnostic_target = pkl.load(fl)
with open(gallery_agnostic_link_path, 'rb') as fl:
gallery_agnostic_link = pkl.load(fl)
with open(gallery_agnostic_text_path, 'r') as fl:
gallery_agnostic_text = json.load(fl)
with open(gallery_specific_chat_path, 'r') as fl:
gallery_specific_chat = json.load(fl)
with open(gallery_agnostic_chat_path, 'r') as fl:
gallery_agnostic_chat = json.load(fl)
gallery_specific_list = []
gallery_specific_target_dict = {}
for i, (s, r, o) in enumerate(gallery_specific_target):
s = id_to_meshid[str(s)]
o = id_to_meshid[str(o)]
k = f'{gallery_specific_link[i][0]}_{gallery_specific_link[i][1]}_{gallery_specific_link[i][2]}_{i}'
if 'sorry' in gallery_specific_text[k]['out'] or 'Sorry' in gallery_specific_text[k]['out']:
continue
target_name = f'{capitalize_the_first_letter(entity_raw_name[s])} - {capitalize_the_first_letter(entity_raw_name[o])}'
if target_name not in gallery_specific_target_dict:
gallery_specific_target_dict[target_name] = i
gallery_specific_list.append(target_name)
gallery_specific_list.sort()
gallery_agnostic_list = []
gallery_agnostic_target_dict = {}
for i, iid in enumerate(gallery_agnostic_target):
target_name = capitalize_the_first_letter(entity_raw_name[id_to_meshid[str(iid)]])
k = f'{gallery_agnostic_link[i][0]}_{gallery_agnostic_link[i][1]}_{gallery_agnostic_link[i][2]}_{i}'
if 'sorry' in gallery_agnostic_text[k]['out'] or 'Sorry' in gallery_agnostic_text[k]['out']:
continue
if target_name not in gallery_agnostic_target_dict:
gallery_agnostic_target_dict[target_name] = i
gallery_agnostic_list.append(target_name)
gallery_agnostic_list.sort()
drug_dict = {}
disease_dict = {}
for k, v in entity_raw_name.items():
#chemical_mesh:c050048
tp = k.split('_')[0]
v = capitalize_the_first_letter(v)
if len(v) <= 4:
continue
if tp == 'chemical':
drug_dict[v] = k
elif tp == 'disease':
disease_dict[v] = k
drug_list = list(drug_dict.keys())
disease_list = list(disease_dict.keys())
drug_list.sort()
disease_list.sort()
init_mask = np.asarray([0] * n_ent).astype('int64')
init_mask = (init_mask == 1)
for k, v in filters.items():
for kk, vv in v.items():
tmp = init_mask.copy()
tmp[np.asarray(vv)] = True
t = torch.ByteTensor(tmp).to(args.device)
filters[k][kk] = t
print('done')
gpt_tokenizer = AutoTokenizer.from_pretrained('microsoft/biogpt')
gpt_tokenizer.pad_token = gpt_tokenizer.eos_token
gpt_model = BioGptForCausalLM.from_pretrained('microsoft/biogpt', pad_token_id=gpt_tokenizer.eos_token_id)
gpt_model.eval()
specific_model = utils.load_model(model_path, args, n_ent, n_rel, args.device)
specific_model.eval()
divide_bound, data_mean, data_std = attack.calculate_edge_bound(data, specific_model, args.device, n_ent)
print('done')
nlp = spacy.load("en_core_web_sm")
bart_model = BartForConditionalGeneration.from_pretrained('GanjinZero/biobart-large')
bart_model.eval()
bart_tokenizer = AutoTokenizer.from_pretrained('GanjinZero/biobart-large')
def tune_chatgpt(draft, attack_data, dpath):
dpath_i = 0
bart_model.to(args.device1)
for i, v in enumerate(draft):
input = v['in'].replace('\n', '')
output = v['out'].replace('\n', '')
s, r, o = attack_data[i]
path_text = dpath[dpath_i].replace('\n', '')
dpath_i += 1
text_s = entity_raw_name[id_to_meshid[s]]
text_o = entity_raw_name[id_to_meshid[o]]
doc = nlp(output)
words= input.split(' ')
tokenized_sens = [sen for sen in doc.sents]
sens = np.array([sen.text for sen in doc.sents])
checkset = set([text_s, text_o])
e_entity = set(['start_entity', 'end_entity'])
for path in path_text.split(' '):
a, b, c = path.split('|')
if a not in e_entity:
checkset.add(a)
if c not in e_entity:
checkset.add(c)
vec = []
l = 0
while(l < len(words)):
bo =False
for j in range(len(words), l, -1): # reversing is important !!!
cc = ' '.join(words[l:j])
if (cc in checkset):
vec += [True] * (j-l)
l = j
bo = True
break
if not bo:
vec.append(False)
l += 1
vec, span = server_utils.find_mini_span(vec, words, checkset)
# vec = np.vectorize(lambda x: x in checkset)(words)
vec[-1] = True
prompt = []
mask_num = 0
for j, bo in enumerate(vec):
if not bo:
mask_num += 1
else:
if mask_num > 0:
# mask_num = mask_num // 3 # span length ~ poisson distribution (lambda = 3)
mask_num = max(mask_num, 1)
mask_num= min(8, mask_num)
prompt += ['<mask>'] * mask_num
prompt.append(words[j])
mask_num = 0
prompt = ' '.join(prompt)
Text = []
Assist = []
for j in range(len(sens)):
Bart_input = list(sens[:j]) + [prompt] +list(sens[j+1:])
assist = list(sens[:j]) + [input] +list(sens[j+1:])
Text.append(' '.join(Bart_input))
Assist.append(' '.join(assist))
for j in range(len(sens)):
Bart_input = server_utils.mask_func(tokenized_sens[:j]) + [input] + server_utils.mask_func(tokenized_sens[j+1:])
assist = list(sens[:j]) + [input] +list(sens[j+1:])
Text.append(' '.join(Bart_input))
Assist.append(' '.join(assist))
batch_size = 8
Outs = []
for l in tqdm(range(0, len(Text), batch_size)):
R = min(len(Text), l + batch_size)
A = bart_tokenizer(Text[l:R],
truncation = True,
padding = True,
max_length = 1024,
return_tensors="pt")
input_ids = A['input_ids'].to(args.device1)
attention_mask = A['attention_mask'].to(args.device1)
aaid = bart_model.generate(input_ids, attention_mask = attention_mask, num_beams = 5, max_length = 1024)
outs = bart_tokenizer.batch_decode(aaid, skip_special_tokens=True, clean_up_tokenization_spaces=False)
Outs += outs
bart_model.to('cpu')
return span, prompt, Outs, Text, Assist
def score_and_select(s, r, o, span , prompt , sen_list, BART_in, Assist, dpath, v):
criterion = CrossEntropyLoss(reduction="none")
text_s = entity_raw_name[id_to_meshid[str(s)]]
text_o = entity_raw_name[id_to_meshid[str(o)]]
sen_list = [server_utils.process(text) for text in sen_list]
path_text = dpath[0].replace('\n', '')
checkset = set([text_s, text_o])
e_entity = set(['start_entity', 'end_entity'])
for path in path_text.split(' '):
a, b, c = path.split('|')
if a not in e_entity:
checkset.add(a)
if c not in e_entity:
checkset.add(c)
input = v['in'].replace('\n', '')
output = v['out'].replace('\n', '')
doc = nlp(output)
gpt_sens = [sen.text for sen in doc.sents]
assert len(gpt_sens) == len(sen_list) // 2
word_sets = []
for sen in gpt_sens:
word_sets.append(set(sen.split(' ')))
def sen_align(word_sets, modified_word_sets):
l = 0
while(l < len(modified_word_sets)):
if len(word_sets[l].intersection(modified_word_sets[l])) > len(word_sets[l]) * 0.8:
l += 1
else:
break
if l == len(modified_word_sets):
return -1, -1, -1, -1
r = l + 1
r1 = None
r2 = None
for pos1 in range(r, len(word_sets)):
for pos2 in range(r, len(modified_word_sets)):
if len(word_sets[pos1].intersection(modified_word_sets[pos2])) > len(word_sets[pos1]) * 0.8:
r1 = pos1
r2 = pos2
break
if r1 is not None:
break
if r1 is None:
r1 = len(word_sets)
r2 = len(modified_word_sets)
return l, r1, l, r2
replace_sen_list = []
boundary = []
assert len(sen_list) % 2 == 0
for j in range(len(sen_list) // 2):
doc = nlp(sen_list[j])
sens = [sen.text for sen in doc.sents]
modified_word_sets = [set(sen.split(' ')) for sen in sens]
l1, r1, l2, r2 = sen_align(word_sets, modified_word_sets)
boundary.append((l1, r1, l2, r2))
if l1 == -1:
replace_sen_list.append(sen_list[j])
continue
check_text = ' '.join(sens[l2: r2])
replace_sen_list.append(' '.join(gpt_sens[:l1] + [check_text] + gpt_sens[r1:]))
sen_list = replace_sen_list + sen_list[len(sen_list) // 2:]
gpt_model.to(args.device1)
sen_list.append(output)
tokens = gpt_tokenizer( sen_list,
truncation = True,
padding = True,
max_length = 1024,
return_tensors="pt")
target_ids = tokens['input_ids'].to(args.device1)
attention_mask = tokens['attention_mask'].to(args.device1)
L = len(sen_list)
ret_log_L = []
for l in tqdm(range(0, L, 5)):
R = min(L, l + 5)
target = target_ids[l:R, :]
attention = attention_mask[l:R, :]
outputs = gpt_model(input_ids = target,
attention_mask = attention,
labels = target)
logits = outputs.logits
shift_logits = logits[..., :-1, :].contiguous()
shift_labels = target[..., 1:].contiguous()
Loss = criterion(shift_logits.view(-1, shift_logits.shape[-1]), shift_labels.view(-1))
Loss = Loss.view(-1, shift_logits.shape[1])
attention = attention[..., 1:].contiguous()
log_Loss = (torch.mean(Loss * attention.float(), dim = 1) / torch.mean(attention.float(), dim = 1))
ret_log_L.append(log_Loss.detach())
log_Loss = torch.cat(ret_log_L, -1).cpu().numpy()
gpt_model.to('cpu')
p = np.argmin(log_Loss)
return sen_list[p]
def generate_template_for_triplet(attack_data):
criterion = CrossEntropyLoss(reduction="none")
gpt_model.to(args.device1)
print('Generating template ...')
GPT_batch_size = 8
single_sentence = []
test_text = []
test_dp = []
test_parse = []
s, r, o = attack_data[0]
dependency_sen_dict = retieve_sentence_through_edgetype[int(r)]['manual']
candidate_sen = []
Dp_path = []
L = len(dependency_sen_dict.keys())
bound = 500 // L
if bound == 0:
bound = 1
for dp_path, sen_list in dependency_sen_dict.items():
if len(sen_list) > bound:
index = np.random.choice(np.array(range(len(sen_list))), bound, replace=False)
sen_list = [sen_list[aa] for aa in index]
ssen_list = []
for aa in range(len(sen_list)):
paper_id, sen_id = sen_list[aa]
if raw_text_sen[paper_id][sen_id]['start_formatted'] == raw_text_sen[paper_id][sen_id]['end_formatted']:
continue
ssen_list.append(sen_list[aa])
sen_list = ssen_list
candidate_sen += sen_list
Dp_path += [dp_path] * len(sen_list)
text_s = entity_raw_name[id_to_meshid[s]]
text_o = entity_raw_name[id_to_meshid[o]]
candidate_text_sen = []
candidate_ori_sen = []
candidate_parse_sen = []
for paper_id, sen_id in candidate_sen:
sen = raw_text_sen[paper_id][sen_id]
text = sen['text']
candidate_ori_sen.append(text)
ss = sen['start_formatted']
oo = sen['end_formatted']
text = text.replace('-LRB-', '(')
text = text.replace('-RRB-', ')')
text = text.replace('-LSB-', '[')
text = text.replace('-RSB-', ']')
text = text.replace('-LCB-', '{')
text = text.replace('-RCB-', '}')
parse_text = text
parse_text = parse_text.replace(ss, text_s.replace(' ', '_'))
parse_text = parse_text.replace(oo, text_o.replace(' ', '_'))
text = text.replace(ss, text_s)
text = text.replace(oo, text_o)
text = text.replace('_', ' ')
candidate_text_sen.append(text)
candidate_parse_sen.append(parse_text)
tokens = gpt_tokenizer( candidate_text_sen,
truncation = True,
padding = True,
max_length = 300,
return_tensors="pt")
target_ids = tokens['input_ids'].to(args.device1)
attention_mask = tokens['attention_mask'].to(args.device1)
L = len(candidate_text_sen)
assert L > 0
ret_log_L = []
for l in tqdm(range(0, L, GPT_batch_size)):
R = min(L, l + GPT_batch_size)
target = target_ids[l:R, :]
attention = attention_mask[l:R, :]
outputs = gpt_model(input_ids = target,
attention_mask = attention,
labels = target)
logits = outputs.logits
shift_logits = logits[..., :-1, :].contiguous()
shift_labels = target[..., 1:].contiguous()
Loss = criterion(shift_logits.view(-1, shift_logits.shape[-1]), shift_labels.view(-1))
Loss = Loss.view(-1, shift_logits.shape[1])
attention = attention[..., 1:].contiguous()
log_Loss = (torch.mean(Loss * attention.float(), dim = 1) / torch.mean(attention.float(), dim = 1))
ret_log_L.append(log_Loss.detach())
ret_log_L = list(torch.cat(ret_log_L, -1).cpu().numpy())
sen_score = list(zip(candidate_text_sen, ret_log_L, candidate_ori_sen, Dp_path, candidate_parse_sen))
sen_score.sort(key = lambda x: x[1])
Len = len(sen_score)
p = 0
if Len > 10:
p = np.random.choice(np.array(range(Len // 10)), 1)[0]
test_text.append(sen_score[p][2])
test_dp.append(sen_score[p][3])
test_parse.append(sen_score[p][4])
single_sentence.append(sen_score[p][0])
gpt_model.to('cpu')
return single_sentence, test_text, test_dp, test_parse
meshids = list(id_to_meshid.values())
cal = {
'chemical' : 0,
'disease' : 0,
'gene' : 0
}
for meshid in meshids:
cal[meshid.split('_')[0]] += 1
def check_reasonable(s, r, o):
train_trip = np.asarray([[s, r, o]])
train_trip = torch.from_numpy(train_trip.astype('int64')).to(device)
edge_loss = get_model_loss_without_softmax(train_trip, specific_model, device).squeeze()
# edge_losse_log_prob = torch.log(F.softmax(-edge_loss, dim = -1))
edge_loss = edge_loss.item()
edge_loss = (edge_loss - data_mean) / data_std
edge_losses_prob = 1 / ( 1 + np.exp(edge_loss - divide_bound) )
bound = 1 - args.reasonable_rate
return (edge_losses_prob > bound), edge_losses_prob
edgeid_to_edgetype = {}
edgeid_to_reversemask = {}
for k, id_list in Parameters.edge_type_to_id.items():
for iid, mask in zip(id_list, Parameters.reverse_mask[k]):
edgeid_to_edgetype[str(iid)] = k
edgeid_to_reversemask[str(iid)] = mask
reverse_tot = 0
G = nx.DiGraph()
for s, r, o in data:
assert id_to_meshid[s].split('_')[0] == edgeid_to_edgetype[r].split('-')[0]
if edgeid_to_reversemask[r] == 1:
reverse_tot += 1
G.add_edge(int(o), int(s))
else:
G.add_edge(int(s), int(o))
print('Page ranking ...')
pagerank_value_1 = nx.pagerank(G, max_iter = 200, tol=1.0e-7)
drug_meshid = []
drug_list = []
for meshid, nm in entity_raw_name.items():
if nm.lower() in drug_term and meshid.split('_')[0] == 'chemical':
drug_meshid.append(meshid)
drug_list.append(capitalize_the_first_letter(nm))
drug_list = list(set(drug_list))
drug_list.sort()
drug_meshid = set(drug_meshid)
pr = list(pagerank_value_1.items())
pr.sort(key = lambda x: x[1])
sorted_rank = { 'chemical' : [],
'gene' : [],
'disease': [],
'merged' : []}
for iid, score in pr:
tp = id_to_meshid[str(iid)].split('_')[0]
if tp == 'chemical':
if id_to_meshid[str(iid)] in drug_meshid:
sorted_rank[tp].append((iid, score))
else:
sorted_rank[tp].append((iid, score))
sorted_rank['merged'].append((iid, score))
llen = len(sorted_rank['merged'])
sorted_rank['merged'] = sorted_rank['merged'][llen * 3 // 4 : ]
def generate_specific_attack_edge(start_entity, end_entity):
if device == torch.device('cpu'):
print('We can just set the malicious link equals to the target link, since the generation of malicious link is too slow on cpu')
return entity_to_id[drug_dict[start_entity]], '10', entity_to_id[disease_dict[end_entity]]
global specific_model
specific_model.to(device)
strat_meshid = drug_dict[start_entity]
end_meshid = disease_dict[end_entity]
start_entity = entity_to_id[strat_meshid]
end_entity = entity_to_id[end_meshid]
target_data = np.array([[start_entity, '10', end_entity]])
neighbors = attack.generate_nghbrs(target_data, edge_nghbrs, args)
ret = f'Generating malicious link for {strat_meshid}_treatment_{end_meshid}', 'Generation malicious text ...'
param_optimizer = list(specific_model.named_parameters())
param_influence = []
for n,p in param_optimizer:
param_influence.append(p)
len_list = []
for v in neighbors.values():
len_list.append(len(v))
mean_len = np.mean(len_list)
attack_trip, score_record = attack.addition_attack(param_influence, args.device, n_rel, data, target_data, neighbors, specific_model, filters, entityid_to_nodetype, args.attack_batch_size, args, load_Record = args.load_existed, divide_bound = divide_bound, data_mean = data_mean, data_std = data_std, cache_intermidiate = False)
s, r, o = attack_trip[0]
specific_model.to('cpu')
return s, r, o
def generate_agnostic_attack_edge(targets):
specific_model.to(device)
attack_edge_list = []
for target in targets:
candidate_list = []
score_list = []
loss_list = []
main_dict = {}
for iid, score in tqdm(sorted_rank['merged']):
a = G.number_of_edges(iid, target) + 1
if a != 1:
continue
b = G.out_degree(iid) + 1
tp = id_to_meshid[str(iid)].split('_')[0]
edge_losses = []
r_list = []
for r in range(len(edgeid_to_edgetype)):
r_tp = edgeid_to_edgetype[str(r)]
if (edgeid_to_reversemask[str(r)] == 0 and r_tp.split('-')[0] == tp and r_tp.split('-')[1] == 'chemical'):
train_trip = np.array([[iid, r, target]])
train_trip = torch.from_numpy(train_trip.astype('int64')).to(device)
edge_loss = get_model_loss_without_softmax(train_trip, specific_model, device).squeeze()
edge_losses.append(edge_loss.unsqueeze(0).detach())
r_list.append(r)
elif(edgeid_to_reversemask[str(r)] == 1 and r_tp.split('-')[0] == 'chemical' and r_tp.split('-')[1] == tp):
train_trip = np.array([[iid, r, target]]) # add batch dim
train_trip = torch.from_numpy(train_trip.astype('int64')).to(device)
edge_loss = get_model_loss_without_softmax(train_trip, specific_model, device).squeeze()
edge_losses.append(edge_loss.unsqueeze(0).detach())
r_list.append(r)
if len(edge_losses)==0:
continue
min_index = torch.argmin(torch.cat(edge_losses, dim = 0))
r = r_list[min_index]
r_tp = edgeid_to_edgetype[str(r)]
old_len = len(candidate_list)
if (edgeid_to_reversemask[str(r)] == 0):
bo, prob = check_reasonable(iid, r, target)
if bo:
candidate_list.append((iid, r, target))
score_list.append(score * a / b)
loss_list.append(edge_losses[min_index].item())
if (edgeid_to_reversemask[str(r)] == 1):
bo, prob = check_reasonable(target, r, iid)
if bo:
candidate_list.append((target, r, iid))
score_list.append(score * a / b)
loss_list.append(edge_losses[min_index].item())
if len(candidate_list) == 0:
if args.added_edge_num == '' or int(args.added_edge_num) == 1:
attack_edge_list.append((-1,-1,-1))
else:
attack_edge_list.append([])
continue
norm_score = np.array(score_list) / np.sum(score_list)
norm_loss = np.exp(-np.array(loss_list)) / np.sum(np.exp(-np.array(loss_list)))
total_score = norm_score * norm_loss
total_score_index = list(zip(range(len(total_score)), total_score))
total_score_index.sort(key = lambda x: x[1], reverse = True)
total_index = np.argsort(total_score)[::-1]
assert total_index[0] == total_score_index[0][0]
# find rank of main index
max_index = np.argmax(total_score)
assert max_index == total_score_index[0][0]
tmp_add = []
add_num = 1
if args.added_edge_num == '' or int(args.added_edge_num) == 1:
attack_edge_list.append(candidate_list[max_index])
else:
add_num = int(args.added_edge_num)
for i in range(add_num):
tmp_add.append(candidate_list[total_score_index[i][0]])
attack_edge_list.append(tmp_add)
specific_model.to('cpu')
return attack_edge_list[0]
def specific_func(start_entity, end_entity, API_key = ''):
try:
args.reasonable_rate = 0.5
s, r, o = generate_specific_attack_edge(start_entity, end_entity)
if int(s) == -1:
return 'All candidate links are filterd out by defender, so no malicious link can be generated', 'No malicious abstract can be generated'
s_name = entity_raw_name[id_to_entity[str(s)]]
r_name = Parameters.edge_id_to_type[int(r)].split(':')[1]
o_name = entity_raw_name[id_to_entity[str(o)]]
attack_data = np.array([[s, r, o]])
path_list = []
with open(f'DiseaseSpecific/generate_abstract/path/random_{args.reasonable_rate}_path.json', 'r') as fl:
for line in fl.readlines():
line.replace('\n', '')
path_list.append(line)
with open(f'DiseaseSpecific/generate_abstract/random_{args.reasonable_rate}_sentence.json', 'r') as fl:
sentence_dict = json.load(fl)
dpath = []
for k, v in sentence_dict.items():
if f'{s}_{r}_{o}' in k:
single_sentence = [v]
dpath = [path_list[int(k.split('_')[-1])]]
break
if len(dpath) == 0:
single_sentence, _, dpath, _ = generate_template_for_triplet(attack_data)
elif not(s_name in single_sentence[0] and o_name in single_sentence[0]):
single_sentence, _, dpath, _ = generate_template_for_triplet(attack_data)
print('Using ChatGPT for generation...')
API_key = API_key.strip()
if API_key != '':
draft = generate_abstract(single_sentence[0], API_key)
else:
draft = generate_abstract(single_sentence[0])
if 'sorry' in draft or 'Sorry' in draft:
return 'All candidate links are filterd out by defender, so no malicious link can be generated', 'No malicious abstract can be generated'
if device != torch.device('cpu'):
print('Using BioBART for tuning...')
span , prompt , sen_list, BART_in, Assist = tune_chatgpt([{'in':single_sentence[0], 'out': draft}], attack_data, dpath)
text = score_and_select(s, r, o, span , prompt , sen_list, BART_in, Assist, dpath, {'in':single_sentence[0], 'out': draft})
else:
text = draft
return f'{capitalize_the_first_letter(s_name)} - {capitalize_the_first_letter(r_name)} - {capitalize_the_first_letter(o_name)}', server_utils.process(text)
# f'The sentence is: {single_sentence[0]}\n The path is: {dpath[0]}'
except:
# return message in error
return 'Error :(', traceback.format_exc()
def agnostic_func(agnostic_entity, API_key = ''):
try:
args.reasonable_rate = 0.7
target_id = entity_to_id[drug_dict[agnostic_entity]]
s = generate_agnostic_attack_edge([int(target_id)])
if len(s) == 0:
return 'All candidate links are filterd out by defender, so no malicious link can be generated', 'No malicious abstract can be generated'
if int(s[0]) == -1:
return 'All candidate links are filterd out by defender, so no malicious link can be generated', 'No malicious abstract can be generated'
s, r, o = str(s[0]), str(s[1]), str(s[2])
s_name = entity_raw_name[id_to_entity[str(s)]]
r_name = Parameters.edge_id_to_type[int(r)].split(':')[1]
o_name = entity_raw_name[id_to_entity[str(o)]]
attack_data = np.array([[s, r, o]])
single_sentence, _, dpath, _ = generate_template_for_triplet(attack_data)
print('Using ChatGPT for generation...')
API_key = API_key.strip()
if API_key != '':
draft = generate_abstract(single_sentence[0], API_key)
else:
draft = generate_abstract(single_sentence[0])
if 'sorry' in draft or 'Sorry' in draft:
return 'All candidate links are filterd out by defender, so no malicious link can be generated', 'No malicious abstract can be generated'
if device != torch.device('cpu'):
print('Using BioBART for tuning...')
span , prompt , sen_list, BART_in, Assist = tune_chatgpt([{'in':single_sentence[0], 'out': draft}], attack_data, dpath)
text = score_and_select(s, r, o, span , prompt , sen_list, BART_in, Assist, dpath, {'in':single_sentence[0], 'out': draft})
else:
text = draft
return f'{capitalize_the_first_letter(s_name)} - {capitalize_the_first_letter(r_name)} - {capitalize_the_first_letter(o_name)}', server_utils.process(text)
except:
# return message in error
return 'Error :(', traceback.format_exc()
def gallery_specific_func(specific_target):
index = gallery_specific_target_dict[specific_target]
s, r, o = gallery_specific_link[index]
s_name = entity_raw_name[id_to_entity[str(s)]]
r_name = Parameters.edge_id_to_type[int(r)].split(':')[1]
o_name = entity_raw_name[id_to_entity[str(o)]]
k = f'{s}_{r}_{o}_{index}'
inn = gallery_specific_text[k]['in']
text = gallery_specific_text[k]['out']
if inn in text:
text = gallery_specific_chat[k]['out']
return f'{capitalize_the_first_letter(s_name)} - {capitalize_the_first_letter(r_name)} - {capitalize_the_first_letter(o_name)}', server_utils.process(text)
def gallery_agnostic_func(agnostic_target):
index = gallery_agnostic_target_dict[agnostic_target]
s, r, o = gallery_agnostic_link[index]
s_name = entity_raw_name[id_to_entity[str(s)]]
r_name = Parameters.edge_id_to_type[int(r)].split(':')[1]
o_name = entity_raw_name[id_to_entity[str(o)]]
k = f'{s}_{r}_{o}_{index}'
inn = gallery_agnostic_text[k]['in']
text = gallery_agnostic_text[k]['out']
if inn in text:
text = gallery_agnostic_chat[k]['out']
return f'{capitalize_the_first_letter(s_name)} - {capitalize_the_first_letter(r_name)} - {capitalize_the_first_letter(o_name)}', server_utils.process(text)
#%%
with gr.Blocks() as demo:
with gr.Column():
gr.Markdown("Poison medical knowledge with Scorpius")
# with gr.Column():
with gr.Row():
# Center
with gr.Column():
gr.Markdown("Select your poisoning target")
with gr.Tab('Gallery'):
with gr.Tab('Target specific'):
specific_target = gr.Dropdown(gallery_specific_list, label="Promoting drug and target disease")
gallery_specific_generation_button = gr.Button('Poison!')
with gr.Tab('Target agnostic'):
agnostic_target = gr.Dropdown(gallery_agnostic_list, label="Promoting drug")
gallery_agnostic_generation_button = gr.Button('Poison!')
with gr.Tab('Poison'):
with gr.Tab('Target specific'):
with gr.Column():
with gr.Row():
start_entity = gr.Dropdown(drug_list, label="Promoting drug")
end_entity = gr.Dropdown(disease_list, label="Target disease")
if device == torch.device('cpu'):
gr.Markdown("Since the project is currently running on the CPU, we directly treat the malicious link as equivalent to the poisoning target, to accelerate the generation process.")
specific_generation_button = gr.Button('Poison!')
gr.Markdown('Please type your openai API key in the textbox below before clicking the **Poison!** button. If the text box is empty, we will use the default API, but the balance of the default API is limited, so the generation may fail. \n We promise that we will not steal your API key in any way. If you still have this concern, please download the source code from **Files**, then use `CUDA_VISIBLE_DEVICES=0 python server.py` to run the offline version.')
API_key_specific = gr.Textbox(label="API key")
with gr.Tab('Target agnostic'):
agnostic_entity = gr.Dropdown(drug_list, label="Promoting drug")
agnostic_generation_button = gr.Button('Poison!')
gr.Markdown('Please type your openai API key in the textbox below before clicking the **Poison!** button. If the text box is empty, we will use the default API, but the balance of the default API is limited, so the generation may fail. \n We promise that we will not steal your API key in any way. If you still have this concern, please download the source code from **Files**, then use `CUDA_VISIBLE_DEVICES=0 python server.py` to run the offline version.')
API_key_agnostic = gr.Textbox(label="API key")
with gr.Column():
gr.Markdown("Generation")
malicisous_link = gr.Textbox(lines=1, label="Malicious link")
# gr.Markdown("Malicious text")
malicious_text = gr.Textbox(label="Malicious text", lines=5)
specific_generation_button.click(specific_func, inputs=[start_entity, end_entity, API_key_specific], outputs=[malicisous_link, malicious_text])
agnostic_generation_button.click(agnostic_func, inputs=[agnostic_entity, API_key_agnostic], outputs=[malicisous_link, malicious_text])
gallery_specific_generation_button.click(gallery_specific_func, inputs=[specific_target], outputs=[malicisous_link, malicious_text])
gallery_agnostic_generation_button.click(gallery_agnostic_func, inputs=[agnostic_target], outputs=[malicisous_link, malicious_text])
# demo.launch(server_name="0.0.0.0", server_port=8000, debug=False)
demo.launch() |