Spaces:
Paused
Paused
File size: 41,521 Bytes
6e476e3 73db451 f9d16d0 39d09a2 6e476e3 5a1417d 73db451 6e476e3 39d09a2 f9d16d0 5a1417d 73db451 f9d16d0 5a1417d 73db451 6e476e3 f9d16d0 39d09a2 5a1417d 6e476e3 4c77d06 73db451 4c77d06 5a1417d 4c77d06 5a1417d 4c77d06 73db451 4c77d06 73db451 4c77d06 73db451 5a1417d 73db451 6e476e3 73db451 6e476e3 73db451 6e476e3 73db451 6e476e3 73db451 6e476e3 73db451 6e476e3 4c77d06 5a1417d 4c77d06 5a1417d 4c77d06 73db451 6e476e3 73db451 6a26e21 73db451 6e476e3 4c77d06 5a1417d 4c77d06 6e476e3 73db451 6e476e3 4c77d06 6e476e3 73db451 6e476e3 4c77d06 5a1417d 4c77d06 5a1417d 4c77d06 5a1417d 4c77d06 5a1417d 73db451 5a1417d 6e476e3 4c77d06 73db451 5a1417d 73db451 5a1417d 73db451 5a1417d 73db451 5a1417d 73db451 5a1417d 6e476e3 73db451 6e476e3 4c77d06 5a1417d 4c77d06 73db451 5a1417d 73db451 5a1417d 73db451 5a1417d 73db451 f9d16d0 73db451 5a1417d 39d09a2 f9d16d0 5a1417d 4c77d06 39d09a2 f9d16d0 5a1417d f9d16d0 39d09a2 5a1417d 4c77d06 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 4c77d06 42f5370 4c77d06 5a1417d 42f5370 4c77d06 42f5370 73db451 4c77d06 5a1417d 4c77d06 5a1417d 4c77d06 f9d16d0 42f5370 f9d16d0 42f5370 4c77d06 5a1417d 42f5370 4c77d06 42f5370 4c77d06 42f5370 f9d16d0 42f5370 f9d16d0 42f5370 4c77d06 5a1417d 42f5370 4c77d06 42f5370 f9d16d0 42f5370 f9d16d0 42f5370 4c77d06 42f5370 4c77d06 42f5370 4c77d06 5a1417d 42f5370 73db451 4c77d06 f9d16d0 4c77d06 73db451 f9d16d0 73db451 f9d16d0 5a1417d f9d16d0 5a1417d 73db451 f9d16d0 5a1417d f9d16d0 5a1417d f9d16d0 5a1417d f9d16d0 5a1417d f9d16d0 5a1417d f9d16d0 5a1417d f9d16d0 5a1417d f9d16d0 5a1417d f9d16d0 42f5370 f9d16d0 42f5370 f9d16d0 39d09a2 f9d16d0 39d09a2 f9d16d0 5a1417d f9d16d0 5a1417d f9d16d0 5a1417d f9d16d0 39d09a2 f9d16d0 73db451 f9d16d0 73db451 5a1417d 39d09a2 4c77d06 5a1417d 4c77d06 5a1417d 4c77d06 42f5370 6e476e3 4c77d06 39d09a2 f9d16d0 39d09a2 f9d16d0 6e476e3 73db451 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 | import gradio as gr
import pandas as pd
import torch
from torch import nn
from transformers import (
BertTokenizer,
BertForSequenceClassification,
TrainingArguments,
Trainer
)
from datasets import Dataset
from sklearn.metrics import (
accuracy_score,
precision_recall_fscore_support,
roc_auc_score,
confusion_matrix
)
import numpy as np
from datetime import datetime
import json
import os
import gc # 用於記憶體清理
# PEFT 相關的 import(LoRA 和 AdaLoRA)
try:
from peft import (
LoraConfig,
AdaLoraConfig,
get_peft_model,
TaskType,
PeftModel
)
PEFT_AVAILABLE = True
except ImportError:
PEFT_AVAILABLE = False
print("⚠️ PEFT 未安裝,LoRA 和 AdaLoRA 功能將不可用")
# 檢查 GPU
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
_MODEL_PATH = None
LAST_TOKENIZER = None
LAST_TUNING_METHOD = None
def evaluate_baseline_bert(eval_dataset, df_clean):
"""
評估原始 BERT(完全沒看過資料)的表現
這部分是從您的格子 5 提取的 baseline 比較邏輯
"""
print("\n" + "=" * 80)
print("評估 Baseline 純 BERT(完全沒看過資料)")
print("=" * 80)
# 載入純 BERT
baseline_tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
baseline_model = BertForSequenceClassification.from_pretrained(
"bert-base-uncased",
num_labels=2
).to(device)
baseline_model.eval()
print(" ⚠️ 這個模型完全沒有使用您的資料訓練")
# 重新處理驗證集
baseline_dataset = Dataset.from_pandas(df_clean[['text', 'label']])
def baseline_preprocess(examples):
return baseline_tokenizer(examples['text'], truncation=True, padding='max_length', max_length=256)
baseline_tokenized = baseline_dataset.map(baseline_preprocess, batched=True)
baseline_split = baseline_tokenized.train_test_split(test_size=0.2, seed=42)
baseline_eval_dataset = baseline_split['test']
# 建立 Baseline Trainer
baseline_trainer_args = TrainingArguments(
output_dir='./temp_baseline',
per_device_eval_batch_size=32,
report_to="none"
)
baseline_trainer = Trainer(
model=baseline_model,
args=baseline_trainer_args,
)
# 評估 Baseline
print("🔄 評估純 BERT...")
predictions_output = baseline_trainer.predict(baseline_eval_dataset)
all_preds = predictions_output.predictions.argmax(-1)
all_labels = predictions_output.label_ids
probs = torch.nn.functional.softmax(torch.tensor(predictions_output.predictions), dim=-1)[:, 1].numpy()
# 計算指標
precision, recall, f1, _ = precision_recall_fscore_support(
all_labels, all_preds, average='binary', pos_label=1, zero_division=0
)
acc = accuracy_score(all_labels, all_preds)
try:
auc = roc_auc_score(all_labels, probs)
except:
auc = 0.0
cm = confusion_matrix(all_labels, all_preds)
if cm.shape == (2, 2):
tn, fp, fn, tp = cm.ravel()
sensitivity = tp / (tp + fn) if (tp + fn) > 0 else 0
specificity = tn / (tn + fp) if (tn + fp) > 0 else 0
else:
sensitivity = specificity = 0
tn = fp = fn = tp = 0
baseline_results = {
'f1': float(f1),
'accuracy': float(acc),
'precision': float(precision),
'recall': float(recall),
'sensitivity': float(sensitivity),
'specificity': float(specificity),
'auc': float(auc),
'tp': int(tp),
'tn': int(tn),
'fp': int(fp),
'fn': int(fn)
}
print("✅ Baseline 評估完成")
return baseline_results
def run_original_code_with_tuning(
file_path,
weight_multiplier,
epochs,
batch_size,
learning_rate,
warmup_steps,
tuning_method,
best_metric,
# LoRA 參數
lora_r,
lora_alpha,
lora_dropout,
lora_modules,
# AdaLoRA 參數
adalora_init_r,
adalora_target_r,
adalora_tinit,
adalora_tfinal,
adalora_delta_t
):
"""
您的原始程式碼 + 不同微調方法的選項 + Baseline 比較
核心邏輯完全不變,只是在模型初始化部分加入條件判斷
"""
global LAST_MODEL_PATH, LAST_TOKENIZER, LAST_TUNING_METHOD
# ==================== 清空記憶體(訓練前) ====================
import gc
torch.cuda.empty_cache()
gc.collect()
print("🧹 記憶體已清空")
# ==================== 您的原始程式碼開始 ====================
# 讀取上傳的檔案
df_original = pd.read_csv(file_path)
df_clean = pd.DataFrame({
'text': df_original['Text'],
'label': df_original['label']
})
df_clean = df_clean.dropna()
print("\n" + "=" * 80)
print(f"乳癌存活預測 BERT Fine-tuning - {tuning_method} 方法")
print("=" * 80)
print(f"開始時間: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
print(f"微調方法: {tuning_method}")
print(f"最佳化指標: {best_metric}")
print("=" * 80)
# 載入 Tokenizer
print("\n📦 載入 BERT Tokenizer...")
tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
print("✅ Tokenizer 載入完成")
# 評估函數 - 完全是您的原始程式碼,不動
def compute_metrics(pred):
labels = pred.label_ids
preds = pred.predictions.argmax(-1)
probs = torch.nn.functional.softmax(torch.tensor(pred.predictions), dim=-1)[:, 1].numpy()
precision, recall, f1, _ = precision_recall_fscore_support(
labels, preds, average='binary', pos_label=1, zero_division=0
)
acc = accuracy_score(labels, preds)
try:
auc = roc_auc_score(labels, probs)
except:
auc = 0.0
cm = confusion_matrix(labels, preds)
if cm.shape == (2, 2):
tn, fp, fn, tp = cm.ravel()
else:
if len(np.unique(preds)) == 1:
if preds[0] == 0:
tn, fp, fn, tp = sum(labels == 0), 0, sum(labels == 1), 0
else:
tn, fp, fn, tp = 0, sum(labels == 0), 0, sum(labels == 1)
else:
tn = fp = fn = tp = 0
sensitivity = tp / (tp + fn) if (tp + fn) > 0 else 0
specificity = tn / (tn + fp) if (tn + fp) > 0 else 0
return {
'accuracy': acc, 'f1': f1, 'precision': precision, 'recall': recall,
'auc': auc, 'sensitivity': sensitivity, 'specificity': specificity,
'tp': int(tp), 'tn': int(tn), 'fp': int(fp), 'fn': int(fn)
}
# ============================================================================
# 步驟 1:準備資料(不做平衡)- 您的原始程式碼
# ============================================================================
print("\n" + "=" * 80)
print("步驟 1:準備資料(保持原始比例)")
print("=" * 80)
print(f"\n原始資料分布:")
print(f" 存活 (0): {sum(df_clean['label']==0)} 筆 ({sum(df_clean['label']==0)/len(df_clean)*100:.1f}%)")
print(f" 死亡 (1): {sum(df_clean['label']==1)} 筆 ({sum(df_clean['label']==1)/len(df_clean)*100:.1f}%)")
ratio = sum(df_clean['label']==0) / sum(df_clean['label']==1)
print(f" 不平衡比例: {ratio:.1f}:1")
# ============================================================================
# 步驟 2:Tokenization - 您的原始程式碼
# ============================================================================
print("\n" + "=" * 80)
print("步驟 2:Tokenization")
print("=" * 80)
dataset = Dataset.from_pandas(df_clean[['text', 'label']])
def preprocess_function(examples):
return tokenizer(examples['text'], truncation=True, padding='max_length', max_length=256)
tokenized_dataset = dataset.map(preprocess_function, batched=True)
train_test_split = tokenized_dataset.train_test_split(test_size=0.2, seed=42)
train_dataset = train_test_split['train']
eval_dataset = train_test_split['test']
print(f"\n✅ 資料集準備完成:")
print(f" 訓練集: {len(train_dataset)} 筆")
print(f" 驗證集: {len(eval_dataset)} 筆")
# ============================================================================
# 步驟 3:設定權重 - 您的原始程式碼
# ============================================================================
print("\n" + "=" * 80)
print(f"步驟 3:設定類別權重({weight_multiplier}x 倍數)")
print("=" * 80)
weight_0 = 1.0
weight_1 = ratio * weight_multiplier
print(f"\n權重設定:")
print(f" 倍數: {weight_multiplier}x")
print(f" 存活類權重: {weight_0:.3f}")
print(f" 死亡類權重: {weight_1:.3f} (= {ratio:.1f} × {weight_multiplier})")
class_weights = torch.tensor([weight_0, weight_1], dtype=torch.float).to(device)
# ============================================================================
# 步驟 4:訓練模型 - 這裡加入不同微調方法的選擇
# ============================================================================
print("\n" + "=" * 80)
print(f"步驟 4:訓練 {tuning_method} BERT 模型")
print("=" * 80)
print(f"\n🔄 初始化模型 ({tuning_method})...")
# 基礎模型載入
model = BertForSequenceClassification.from_pretrained(
"bert-base-uncased", num_labels=2, problem_type="single_label_classification"
)
# 根據選擇的微調方法設定模型
if tuning_method == "Full Fine-tuning":
# 您的原始方法 - 完全不動
model = model.to(device)
print("✅ 使用完整 Fine-tuning(所有參數可訓練)")
trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
all_params = sum(p.numel() for p in model.parameters())
print(f" 可訓練參數: {trainable_params:,} / {all_params:,} ({100 * trainable_params / all_params:.2f}%)")
elif tuning_method == "LoRA" and PEFT_AVAILABLE:
# LoRA 設定
target_modules = lora_modules.split(",") if lora_modules else ["query", "value"]
target_modules = [m.strip() for m in target_modules]
peft_config = LoraConfig(
task_type=TaskType.SEQ_CLS,
r=int(lora_r),
lora_alpha=int(lora_alpha),
lora_dropout=float(lora_dropout),
target_modules=target_modules
)
model = get_peft_model(model, peft_config)
model = model.to(device)
print("✅ 使用 LoRA 微調")
print(f" LoRA rank (r): {lora_r}")
print(f" LoRA alpha: {lora_alpha}")
print(f" LoRA dropout: {lora_dropout}")
print(f" 目標模組: {target_modules}")
model.print_trainable_parameters()
elif tuning_method == "AdaLoRA" and PEFT_AVAILABLE:
# AdaLoRA 設定
target_modules = lora_modules.split(",") if lora_modules else ["query", "value"]
target_modules = [m.strip() for m in target_modules]
peft_config = AdaLoraConfig(
task_type=TaskType.SEQ_CLS,
init_r=int(adalora_init_r),
target_r=int(adalora_target_r),
tinit=int(adalora_tinit),
tfinal=int(adalora_tfinal),
deltaT=int(adalora_delta_t),
lora_alpha=int(lora_alpha),
lora_dropout=float(lora_dropout),
target_modules=target_modules
)
model = get_peft_model(model, peft_config)
model = model.to(device)
print("✅ 使用 AdaLoRA 微調")
print(f" 初始 rank: {adalora_init_r}")
print(f" 目標 rank: {adalora_target_r}")
print(f" Tinit: {adalora_tinit}, Tfinal: {adalora_tfinal}, DeltaT: {adalora_delta_t}")
model.print_trainable_parameters()
else:
# 預設使用 Full Fine-tuning
model = model.to(device)
print("⚠️ PEFT 未安裝或方法無效,使用 Full Fine-tuning")
# 自訂 Trainer(使用權重)- 您的原始程式碼
class WeightedTrainer(Trainer):
def compute_loss(self, model, inputs, return_outputs=False):
labels = inputs.pop("labels")
outputs = model(**inputs)
loss_fct = nn.CrossEntropyLoss(weight=class_weights)
loss = loss_fct(outputs.logits.view(-1, 2), labels.view(-1))
return (loss, outputs) if return_outputs else loss
# 訓練設定 - 根據選擇的最佳指標調整
metric_map = {
"f1": "f1",
"accuracy": "accuracy",
"precision": "precision",
"recall": "recall",
"sensitivity": "sensitivity",
"specificity": "specificity",
"auc": "auc"
}
training_args = TrainingArguments(
output_dir='./results_weight',
num_train_epochs=epochs,
per_device_train_batch_size=batch_size,
per_device_eval_batch_size=batch_size*2,
warmup_steps=warmup_steps,
weight_decay=0.01,
learning_rate=learning_rate,
logging_steps=50,
evaluation_strategy="epoch",
save_strategy="epoch",
load_best_model_at_end=True,
metric_for_best_model=metric_map.get(best_metric, "f1"), # 使用選擇的指標
report_to="none",
greater_is_better=True # 所有指標都是越高越好
)
trainer = WeightedTrainer(
model=model, args=training_args,
train_dataset=train_dataset, eval_dataset=eval_dataset,
compute_metrics=compute_metrics
)
print(f"\n🚀 開始訓練({epochs} epochs)...")
print(f" 最佳化指標: {best_metric}")
print("-" * 80)
trainer.train()
print("\n✅ 模型訓練完成!")
# 評估模型
print("\n📊 評估模型...")
results = trainer.evaluate()
print(f"\n{tuning_method} BERT ({weight_multiplier}x 權重) 表現:")
print(f" F1 Score: {results['eval_f1']:.4f}")
print(f" Accuracy: {results['eval_accuracy']:.4f}")
print(f" Precision: {results['eval_precision']:.4f}")
print(f" Recall: {results['eval_recall']:.4f}")
print(f" Sensitivity: {results['eval_sensitivity']:.4f}")
print(f" Specificity: {results['eval_specificity']:.4f}")
print(f" AUC: {results['eval_auc']:.4f}")
print(f" 混淆矩陣: Tp={results['eval_tp']}, Tn={results['eval_tn']}, "
f"Fp={results['eval_fp']}, Fn={results['eval_fn']}")
# ============================================================================
# 步驟 5:Baseline 比較(純 BERT)- 從您的原始程式碼
# ============================================================================
print("\n" + "=" * 80)
print("步驟 5:Baseline 比較 - 純 BERT(完全沒看過資料)")
print("=" * 80)
baseline_results = evaluate_baseline_bert(eval_dataset, df_clean)
# ============================================================================
# 步驟 6:比較結果 - 從您的原始程式碼
# ============================================================================
print("\n" + "=" * 80)
print(f"📊 【對比結果】純 BERT vs {tuning_method} BERT")
print("=" * 80)
print("\n📋 詳細比較表:")
print("-" * 100)
print(f"{'指標':<15} {'純 BERT':<20} {tuning_method:<20} {'改善幅度':<20}")
print("-" * 100)
metrics_to_compare = [
('F1 Score', 'f1', 'eval_f1'),
('Accuracy', 'accuracy', 'eval_accuracy'),
('Precision', 'precision', 'eval_precision'),
('Recall', 'recall', 'eval_recall'),
('Sensitivity', 'sensitivity', 'eval_sensitivity'),
('Specificity', 'specificity', 'eval_specificity'),
('AUC', 'auc', 'eval_auc')
]
for name, baseline_key, finetuned_key in metrics_to_compare:
baseline_val = baseline_results[baseline_key]
finetuned_val = results[finetuned_key]
improvement = ((finetuned_val - baseline_val) / baseline_val * 100) if baseline_val > 0 else 0
print(f"{name:<15} {baseline_val:<20.4f} {finetuned_val:<20.4f} {improvement:>+18.1f}%")
print("-" * 100)
# 儲存模型
save_dir = f'./breast_cancer_bert_{tuning_method.lower().replace(" ", "_")}'
if tuning_method in ["LoRA", "AdaLoRA"] and PEFT_AVAILABLE:
# PEFT 模型儲存方式
model.save_pretrained(save_dir)
tokenizer.save_pretrained(save_dir)
else:
# 一般模型儲存方式
model.save_pretrained(save_dir)
tokenizer.save_pretrained(save_dir)
# 儲存模型資訊到 JSON 檔案(用於預測頁面選擇)
model_info = {
'model_path': save_dir,
'tuning_method': tuning_method,
'best_metric': best_metric,
'best_metric_value': float(results[f'eval_{metric_map.get(best_metric, "f1")}']),
'timestamp': datetime.now().strftime('%Y-%m-%d %H:%M:%S'),
'weight_multiplier': weight_multiplier,
'epochs': epochs
}
# 讀取現有的模型列表
models_list_file = './saved_models_list.json'
if os.path.exists(models_list_file):
with open(models_list_file, 'r') as f:
models_list = json.load(f)
else:
models_list = []
# 加入新模型資訊
models_list.append(model_info)
# 儲存更新後的列表
with open(models_list_file, 'w') as f:
json.dump(models_list, f, indent=2)
# 儲存到全域變數供預測使用
LAST_MODEL_PATH = save_dir
LAST_TOKENIZER = tokenizer
LAST_TUNING_METHOD = tuning_method
print(f"\n💾 模型已儲存至: {save_dir}")
print("\n" + "=" * 80)
print("🎉 訓練完成!")
print("=" * 80)
print(f"完成時間: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
# ==================== 清空記憶體(訓練後) ====================
del model
del trainer
torch.cuda.empty_cache()
gc.collect()
print("🧹 訓練後記憶體已清空")
# 加入所有資訊到結果中
results['tuning_method'] = tuning_method
results['best_metric'] = best_metric
results['best_metric_value'] = results[f'eval_{metric_map.get(best_metric, "f1")}']
results['baseline_results'] = baseline_results
results['model_path'] = save_dir
return results
def predict_text(model_choice, text_input):
"""
預測功能 - 支援選擇已訓練的模型,並同時顯示未微調和微調的預測結果
"""
if not text_input or text_input.strip() == "":
return "請輸入文本", "請輸入文本"
try:
# ==================== 未微調的 BERT 預測 ====================
print("\n使用未微調 BERT 預測...")
baseline_tokenizer = BertTokenizer.from_pretrained("bert-base-uncased")
baseline_model = BertForSequenceClassification.from_pretrained(
"bert-base-uncased",
num_labels=2
).to(device)
baseline_model.eval()
# Tokenize 輸入(未微調)
baseline_inputs = baseline_tokenizer(
text_input,
truncation=True,
padding='max_length',
max_length=256,
return_tensors='pt'
).to(device)
# 預測(未微調)
with torch.no_grad():
baseline_outputs = baseline_model(**baseline_inputs)
baseline_probs = torch.nn.functional.softmax(baseline_outputs.logits, dim=-1)
baseline_pred_class = baseline_probs.argmax(-1).item()
baseline_confidence = baseline_probs[0][baseline_pred_class].item()
baseline_result = "存活" if baseline_pred_class == 0 else "死亡"
baseline_prob_survive = baseline_probs[0][0].item()
baseline_prob_death = baseline_probs[0][1].item()
baseline_output = f"""
# 🔵 未微調 BERT 預測結果
## 預測類別: **{baseline_result}**
## 信心度: **{baseline_confidence:.1%}**
## 機率分布:
- 🟢 **存活機率**: {baseline_prob_survive:.2%}
- 🔴 **死亡機率**: {baseline_prob_death:.2%}
---
**說明**: 此為原始 BERT 模型,未經任何領域資料訓練
"""
# 清空記憶體
del baseline_model
del baseline_tokenizer
torch.cuda.empty_cache()
# ==================== 微調後的 BERT 預測 ====================
if model_choice == "請先訓練模型":
finetuned_output = """
# 🟢 微調 BERT 預測結果
❌ 尚未訓練任何模型,請先在「模型訓練」頁面訓練模型
"""
return baseline_output, finetuned_output
# 解析選擇的模型路徑
model_path = model_choice.split(" | ")[0].replace("路徑: ", "")
# 從 JSON 讀取模型資訊
with open('./saved_models_list.json', 'r') as f:
models_list = json.load(f)
selected_model_info = None
for model_info in models_list:
if model_info['model_path'] == model_path:
selected_model_info = model_info
break
if selected_model_info is None:
finetuned_output = f"""
# 🟢 微調 BERT 預測結果
❌ 找不到模型:{model_path}
"""
return baseline_output, finetuned_output
print(f"\n使用微調模型: {model_path}")
# 載入 tokenizer
finetuned_tokenizer = BertTokenizer.from_pretrained(model_path)
# 載入模型
tuning_method = selected_model_info['tuning_method']
if tuning_method in ["LoRA", "AdaLoRA"] and PEFT_AVAILABLE:
# 載入 PEFT 模型
base_model = BertForSequenceClassification.from_pretrained("bert-base-uncased", num_labels=2)
finetuned_model = PeftModel.from_pretrained(base_model, model_path)
finetuned_model = finetuned_model.to(device)
else:
# 載入一般模型
finetuned_model = BertForSequenceClassification.from_pretrained(model_path).to(device)
finetuned_model.eval()
# Tokenize 輸入(微調)
finetuned_inputs = finetuned_tokenizer(
text_input,
truncation=True,
padding='max_length',
max_length=256,
return_tensors='pt'
).to(device)
# 預測(微調)
with torch.no_grad():
finetuned_outputs = finetuned_model(**finetuned_inputs)
finetuned_probs = torch.nn.functional.softmax(finetuned_outputs.logits, dim=-1)
finetuned_pred_class = finetuned_probs.argmax(-1).item()
finetuned_confidence = finetuned_probs[0][finetuned_pred_class].item()
finetuned_result = "存活" if finetuned_pred_class == 0 else "死亡"
finetuned_prob_survive = finetuned_probs[0][0].item()
finetuned_prob_death = finetuned_probs[0][1].item()
finetuned_output = f"""
# 🟢 微調 BERT 預測結果
## 預測類別: **{finetuned_result}**
## 信心度: **{finetuned_confidence:.1%}**
## 機率分布:
- 🟢 **存活機率**: {finetuned_prob_survive:.2%}
- 🔴 **死亡機率**: {finetuned_prob_death:.2%}
---
### 模型資訊:
- **微調方法**: {selected_model_info['tuning_method']}
- **最佳化指標**: {selected_model_info['best_metric']}
- **訓練時間**: {selected_model_info['timestamp']}
- **模型路徑**: {model_path}
---
**注意**: 此預測僅供參考,實際醫療決策應由專業醫師判斷。
"""
# 清空記憶體
del finetuned_model
del finetuned_tokenizer
torch.cuda.empty_cache()
return baseline_output, finetuned_output
except Exception as e:
import traceback
error_msg = f"❌ 預測錯誤:{str(e)}\n\n詳細錯誤訊息:\n{traceback.format_exc()}"
return error_msg, error_msg
def get_available_models():
"""
取得所有已訓練的模型列表
"""
models_list_file = './saved_models_list.json'
if not os.path.exists(models_list_file):
return ["請先訓練模型"]
with open(models_list_file, 'r') as f:
models_list = json.load(f)
if len(models_list) == 0:
return ["請先訓練模型"]
# 格式化模型選項
model_choices = []
for i, model_info in enumerate(models_list, 1):
choice = f"路徑: {model_info['model_path']} | 方法: {model_info['tuning_method']} | 時間: {model_info['timestamp']}"
model_choices.append(choice)
return model_choices
# ============================================================================
# Gradio 介面部分 - 修改輸出為三個格子
# ============================================================================
def train_wrapper(
file,
tuning_method,
weight_mult,
epochs,
batch_size,
lr,
warmup,
best_metric,
lora_r,
lora_alpha,
lora_dropout,
lora_modules,
adalora_init_r,
adalora_target_r,
adalora_tinit,
adalora_tfinal,
adalora_delta_t
):
"""包裝函數,處理 Gradio 的輸入輸出 - 分成三格顯示"""
if file is None:
return "請上傳 CSV 檔案", "", ""
try:
# 呼叫訓練函數
results = run_original_code_with_tuning(
file_path=file.name,
weight_multiplier=weight_mult,
epochs=int(epochs),
batch_size=int(batch_size),
learning_rate=lr,
warmup_steps=int(warmup),
tuning_method=tuning_method,
best_metric=best_metric,
lora_r=lora_r,
lora_alpha=lora_alpha,
lora_dropout=lora_dropout,
lora_modules=lora_modules,
adalora_init_r=adalora_init_r,
adalora_target_r=adalora_target_r,
adalora_tinit=adalora_tinit,
adalora_tfinal=adalora_tfinal,
adalora_delta_t=adalora_delta_t
)
# 取得 baseline 結果
baseline_results = results['baseline_results']
# ==================== 格式化輸出:分成三個部分 ====================
# 第一格:資料資訊 (最上面一大格)
data_info = f"""
# 📊 資料資訊
## 🔧 訓練配置
- **微調方法**: {results['tuning_method']}
- **最佳化指標**: {results['best_metric']}
- **最佳指標值**: {results['best_metric_value']:.4f}
## ⚙️ 訓練參數
- **權重倍數**: {weight_mult}x
- **訓練輪數**: {epochs}
- **批次大小**: {batch_size}
- **學習率**: {lr}
- **Warmup Steps**: {warmup}
✅ 訓練完成!模型已儲存,可在「預測」頁面使用!
"""
# 第二格:純 BERT (未微調) - 中間左邊
baseline_output = f"""
# 🔵 純 BERT (Baseline)
## 未經訓練
### 📈 評估指標
| 指標 | 數值 |
|------|------|
| **F1 Score** | {baseline_results['f1']:.4f} |
| **Accuracy** | {baseline_results['accuracy']:.4f} |
| **Precision** | {baseline_results['precision']:.4f} |
| **Recall** | {baseline_results['recall']:.4f} |
| **Sensitivity** | {baseline_results['sensitivity']:.4f} |
| **Specificity** | {baseline_results['specificity']:.4f} |
| **AUC** | {baseline_results['auc']:.4f} |
### 📈 混淆矩陣
| | 預測:存活 | 預測:死亡 |
|---|-----------|-----------|
| **實際:存活** | TN={baseline_results['tn']} | FP={baseline_results['fp']} |
| **實際:死亡** | FN={baseline_results['fn']} | TP={baseline_results['tp']} |
"""
# 第三格:經微調 BERT - 中間右邊
finetuned_output = f"""
# 🟢 經微調 BERT
## {results['tuning_method']}
### 📈 評估指標
| 指標 | 數值 |
|------|------|
| **F1 Score** | {results['eval_f1']:.4f} |
| **Accuracy** | {results['eval_accuracy']:.4f} |
| **Precision** | {results['eval_precision']:.4f} |
| **Recall** | {results['eval_recall']:.4f} |
| **Sensitivity** | {results['eval_sensitivity']:.4f} |
| **Specificity** | {results['eval_specificity']:.4f} |
| **AUC** | {results['eval_auc']:.4f} |
### 📈 混淆矩陣
| | 預測:存活 | 預測:死亡 |
|---|-----------|-----------|
| **實際:存活** | TN={results['eval_tn']} | FP={results['eval_fp']} |
| **實際:死亡** | FN={results['eval_fn']} | TP={results['eval_tp']} |
"""
return data_info, baseline_output, finetuned_output
except Exception as e:
import traceback
error_msg = f"❌ 錯誤:{str(e)}\n\n詳細錯誤訊息:\n{traceback.format_exc()}"
return error_msg, "", ""
# 建立 Gradio 介面
with gr.Blocks(title="BERT 完整訓練與預測平台", theme=gr.themes.Soft()) as demo:
gr.Markdown("""
# 🏥 BERT 乳癌存活預測 - 完整訓練與預測平台
### 🌟 功能特色:
- 🎯 支援三種微調方法:Full Fine-tuning、LoRA、AdaLoRA
- 📊 自動比較有/無微調的表現差異
- 🎨 可選擇最佳化指標(F1、Accuracy、Precision、Recall 等)
- 🔮 訓練後可直接預測新病例
- 💾 自動儲存最佳模型
""")
with gr.Tab("🎯 模型訓練"):
with gr.Row():
with gr.Column(scale=1):
gr.Markdown("### 📤 資料上傳")
file_input = gr.File(
label="上傳 CSV 檔案",
file_types=[".csv"]
)
gr.Markdown("### 🔧 微調方法選擇")
tuning_method = gr.Radio(
choices=["Full Fine-tuning", "LoRA", "AdaLoRA"],
value="Full Fine-tuning",
label="選擇微調方法",
info="不同的參數效率微調方法"
)
gr.Markdown("### 🎯 最佳模型選擇")
best_metric = gr.Dropdown(
choices=["f1", "accuracy", "precision", "recall", "sensitivity", "specificity", "auc"],
value="f1",
label="選擇最佳化指標",
info="模型會根據此指標選擇最佳檢查點,結果會特別顯示此指標"
)
gr.Markdown("### ⚙️ 基本訓練參數")
weight_slider = gr.Slider(
minimum=0.1,
maximum=2.0,
value=0.8,
step=0.1,
label="權重倍數",
info="調整死亡類別的權重"
)
epochs_input = gr.Number(
value=8,
label="訓練輪數 (Epochs)"
)
batch_size_input = gr.Number(
value=16,
label="批次大小 (Batch Size)"
)
lr_input = gr.Number(
value=2e-5,
label="學習率 (Learning Rate)"
)
warmup_input = gr.Number(
value=200,
label="Warmup Steps"
)
# LoRA 特定參數(預設隱藏)
with gr.Column(visible=False) as lora_params:
gr.Markdown("### 🔷 LoRA 參數")
lora_r = gr.Slider(
minimum=4,
maximum=64,
value=16,
step=4,
label="LoRA Rank (r)",
info="低秩分解的秩"
)
lora_alpha = gr.Slider(
minimum=8,
maximum=128,
value=32,
step=8,
label="LoRA Alpha",
info="LoRA 縮放參數"
)
lora_dropout = gr.Slider(
minimum=0.0,
maximum=0.5,
value=0.1,
step=0.05,
label="LoRA Dropout"
)
lora_modules = gr.Textbox(
value="query,value",
label="目標模組",
info="用逗號分隔"
)
# AdaLoRA 特定參數(預設隱藏)
with gr.Column(visible=False) as adalora_params:
gr.Markdown("### 🔶 AdaLoRA 參數")
adalora_init_r = gr.Slider(
minimum=4,
maximum=64,
value=12,
step=4,
label="初始 Rank"
)
adalora_target_r = gr.Slider(
minimum=4,
maximum=64,
value=8,
step=4,
label="目標 Rank"
)
adalora_tinit = gr.Number(value=0, label="Tinit")
adalora_tfinal = gr.Number(value=0, label="Tfinal")
adalora_delta_t = gr.Number(value=1, label="Delta T")
train_button = gr.Button(
"🚀 開始訓練",
variant="primary",
size="lg"
)
with gr.Column(scale=2):
gr.Markdown("### 📊 訓練結果與比較")
# 第一格:資料資訊(最上面一大格)
data_info_output = gr.Markdown(
value="### 等待訓練...\n\n訓練完成後會顯示資料資訊和訓練配置",
label="資料資訊"
)
# 第二和第三格:並排顯示(中間)
with gr.Row():
# 第二格:純 BERT (左邊)
baseline_output = gr.Markdown(
value="### 純 BERT (未微調)\n等待訓練完成...",
label="純 BERT"
)
# 第三格:經微調 BERT (右邊)
finetuned_output = gr.Markdown(
value="### 經微調 BERT\n等待訓練完成...",
label="經微調 BERT"
)
with gr.Tab("🔮 模型預測"):
gr.Markdown("""
### 使用訓練好的模型進行預測
選擇已訓練的模型,輸入病歷文本進行預測。會同時顯示未微調和微調模型的預測結果以供比較。
""")
with gr.Row():
with gr.Column():
# 模型選擇下拉選單
model_dropdown = gr.Dropdown(
label="選擇模型",
choices=["請先訓練模型"],
value="請先訓練模型",
info="選擇要使用的已訓練模型"
)
refresh_button = gr.Button(
"🔄 重新整理模型列表",
size="sm"
)
text_input = gr.Textbox(
label="輸入病歷文本",
placeholder="請輸入患者的病歷描述(英文)...",
lines=10
)
gr.Examples(
examples=[
["Patient is a 45-year-old female with stage II breast cancer, ER+/PR+/HER2-, underwent mastectomy and chemotherapy."],
["65-year-old woman diagnosed with triple-negative breast cancer, stage III, with lymph node involvement."],
["Early stage breast cancer detected, patient is 38 years old, no family history, scheduled for lumpectomy."],
["Patient with advanced metastatic breast cancer, multiple organ involvement, poor prognosis."],
["Young patient, BRCA1 positive, preventive double mastectomy performed, good recovery."]
],
inputs=text_input,
label="範例文本(點擊使用)"
)
predict_button = gr.Button(
"🔮 開始預測",
variant="primary",
size="lg"
)
with gr.Column():
gr.Markdown("### 預測結果比較")
# 上框:未微調 BERT 預測結果
baseline_prediction_output = gr.Markdown(
label="未微調 BERT",
value="等待預測..."
)
# 下框:微調 BERT 預測結果
finetuned_prediction_output = gr.Markdown(
label="微調 BERT",
value="等待預測..."
)
with gr.Tab("📖 使用說明"):
gr.Markdown("""
## 🔧 微調方法說明
| 方法 | 訓練速度 | 記憶體 | 效果 | 適用場景 |
|------|---------|--------|------|----------|
| **Full Fine-tuning** | 1x (基準) | 高 | 最佳 | 資源充足,要最佳效果 |
| **LoRA** | 3-5x 快 | 低 | 良好 | 資源有限,快速實驗 |
| **AdaLoRA** | 3-5x 快 | 低 | 良好 | 自動調整,平衡效果 |
## 📊 指標說明
- **F1 Score**: 精確率和召回率的調和平均,平衡指標
- **Accuracy**: 整體準確率
- **Precision**: 預測為死亡中的準確率
- **Recall/Sensitivity**: 實際死亡中被正確識別的比例
- **Specificity**: 實際存活中被正確識別的比例
- **AUC**: ROC 曲線下面積,整體分類能力
## 💡 使用建議
1. **資料不平衡嚴重**:增加權重倍數,使用 Recall 作為最佳化指標
2. **避免誤診**:使用 Precision 作為最佳化指標
3. **整體平衡**:使用 F1 Score 作為最佳化指標
4. **快速實驗**:使用 LoRA,減少 epochs
5. **最佳效果**:使用 Full Fine-tuning,8-10 epochs
## ⚠️ 注意事項
- 訓練時間依資料量和方法而定(5-20 分鐘)
- 建議至少 100 筆訓練資料
- GPU 會顯著加速訓練
- 預測結果僅供參考,實際醫療決策應由專業醫師判斷
""")
# 根據選擇的微調方法顯示/隱藏相應參數
def update_params_visibility(method):
if method == "LoRA":
return gr.update(visible=True), gr.update(visible=False)
elif method == "AdaLoRA":
return gr.update(visible=True), gr.update(visible=True)
else:
return gr.update(visible=False), gr.update(visible=False)
tuning_method.change(
fn=update_params_visibility,
inputs=[tuning_method],
outputs=[lora_params, adalora_params]
)
# 設定訓練按鈕動作 - 注意這裡改為三個輸出
train_button.click(
fn=train_wrapper,
inputs=[
file_input,
tuning_method,
weight_slider,
epochs_input,
batch_size_input,
lr_input,
warmup_input,
best_metric,
# LoRA 參數
lora_r,
lora_alpha,
lora_dropout,
lora_modules,
# AdaLoRA 參數
adalora_init_r,
adalora_target_r,
adalora_tinit,
adalora_tfinal,
adalora_delta_t
],
outputs=[data_info_output, baseline_output, finetuned_output] # 三個輸出
)
# 重新整理模型列表按鈕
def refresh_models():
return gr.update(choices=get_available_models(), value=get_available_models()[0])
refresh_button.click(
fn=refresh_models,
inputs=[],
outputs=[model_dropdown]
)
# 預測按鈕動作 - 兩個輸出:未微調和微調
predict_button.click(
fn=predict_text,
inputs=[model_dropdown, text_input],
outputs=[baseline_prediction_output, finetuned_prediction_output]
)
if __name__ == "__main__":
demo.launch() |