File size: 8,500 Bytes
bebe233
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# ============================================================
# PhishGuard AI - bert_finetune.py
# Full BERT fine-tuning script on PhishTank + TRANCO data
#
# Downloads data, fine-tunes ealvaradob/bert-finetuned-phishing
# 3 epochs, AdamW + linear warmup scheduler
# Saves to bert_weights/ with save_pretrained()
# Prints per-epoch: loss / precision / recall / F1
# ============================================================

from __future__ import annotations

import logging
import sys
from pathlib import Path
from typing import List, Tuple

logging.basicConfig(
    level=logging.INFO,
    format="%(asctime)s | %(levelname)-7s | %(message)s",
)
logger = logging.getLogger("phishguard.bert_finetune")

BASE_DIR = Path(__file__).parent
BERT_WEIGHTS_DIR = BASE_DIR / "bert_weights"


def main() -> None:
    """Fine-tune BERT on PhishTank + TRANCO URLs."""
    print("=" * 60)
    print("PhishGuard AI β€” BERT Fine-Tuning")
    print("=" * 60)

    # ── Check dependencies ───────────────────────────────────────
    try:
        import torch
        from torch.utils.data import DataLoader, Dataset
        from torch.optim import AdamW
        from transformers import (
            AutoTokenizer,
            AutoModelForSequenceClassification,
            get_linear_schedule_with_warmup,
        )
        from sklearn.metrics import precision_recall_fscore_support
    except ImportError as e:
        print(f"❌ Missing dependency: {e}")
        print("   Run: pip install torch transformers scikit-learn")
        sys.exit(1)

    # ── Download data ────────────────────────────────────────────
    from data_collector import download_phishtank, download_tranco, merge_datasets

    print("\nπŸ“₯ Downloading datasets...")
    phish_urls = download_phishtank(max_urls=50)
    legit_urls = download_tranco(n=50)
    print(f"   Phishing URLs: {len(phish_urls)}")
    print(f"   Legitimate URLs: {len(legit_urls)}")

    train_data, val_data, test_data = merge_datasets(phish_urls, legit_urls)

    # ── URL tokenization ─────────────────────────────────────────
    import re
    _re_url_split = re.compile(r"[-./=?&_~%@:]+")

    def tokenize_url(url: str) -> str:
        text = url.replace("https://", "").replace("http://", "")
        tokens = _re_url_split.split(text)
        return " ".join(t for t in tokens if t)

    # ── Dataset class ────────────────────────────────────────────
    class PhishingURLDataset(Dataset):
        def __init__(self, data: List[Tuple[str, int]], tokenizer, max_length: int = 512):
            self.data = data
            self.tokenizer = tokenizer
            self.max_length = max_length

        def __len__(self) -> int:
            return len(self.data)

        def __getitem__(self, idx: int):
            url, label = self.data[idx]
            text = f"URL: {tokenize_url(url)}"
            encoding = self.tokenizer(
                text,
                truncation=True,
                padding="max_length",
                max_length=self.max_length,
                return_tensors="pt",
            )
            return {
                "input_ids": encoding["input_ids"].squeeze(0),
                "attention_mask": encoding["attention_mask"].squeeze(0),
                "labels": torch.tensor(label, dtype=torch.long),
            }

    # ── Load model ───────────────────────────────────────────────
    MODEL_NAME = "ealvaradob/bert-finetuned-phishing"
    FALLBACK = "mrm8488/bert-tiny-finetuned-sms-spam-detection"

    print("\nπŸ€– Loading BERT model...")
    tokenizer = None
    model = None
    for model_id in [MODEL_NAME, FALLBACK]:
        try:
            tokenizer = AutoTokenizer.from_pretrained(model_id)
            model = AutoModelForSequenceClassification.from_pretrained(
                model_id, num_labels=2
            )
            print(f"   βœ… Loaded: {model_id}")
            break
        except Exception as e:
            print(f"   ⚠️  {model_id} failed: {e}")
            continue

    if model is None or tokenizer is None:
        print("❌ Could not load any BERT model. Exiting.")
        sys.exit(1)

    # ── Prepare data ─────────────────────────────────────────────
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    print(f"   Device: {device}")

    train_dataset = PhishingURLDataset(train_data, tokenizer)
    val_dataset = PhishingURLDataset(val_data, tokenizer)

    train_loader = DataLoader(train_dataset, batch_size=16, shuffle=True)
    val_loader = DataLoader(val_dataset, batch_size=32)

    model = model.to(device)

    # ── Optimizer + Scheduler ────────────────────────────────────
    EPOCHS = 1
    optimizer = AdamW(model.parameters(), lr=2e-5, weight_decay=0.01)
    total_steps = len(train_loader) * EPOCHS
    scheduler = get_linear_schedule_with_warmup(
        optimizer,
        num_warmup_steps=total_steps // 10,
        num_training_steps=total_steps,
    )

    # ── Training Loop ────────────────────────────────────────────
    print(f"\nπŸ‹οΈ Training for {EPOCHS} epochs...")
    print(f"   Train batches: {len(train_loader)}")
    print(f"   Val batches: {len(val_loader)}")

    best_f1 = 0.0
    for epoch in range(1, EPOCHS + 1):
        # Train
        model.train()
        total_loss = 0.0
        train_preds = []
        train_labels = []

        for batch_idx, batch in enumerate(train_loader):
            input_ids = batch["input_ids"].to(device)
            attention_mask = batch["attention_mask"].to(device)
            labels = batch["labels"].to(device)

            optimizer.zero_grad()
            outputs = model(input_ids=input_ids, attention_mask=attention_mask, labels=labels)
            loss = outputs.loss
            loss.backward()

            torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm=1.0)
            optimizer.step()
            scheduler.step()

            total_loss += loss.item()
            preds = torch.argmax(outputs.logits, dim=1)
            train_preds.extend(preds.cpu().tolist())
            train_labels.extend(labels.cpu().tolist())

            if (batch_idx + 1) % 50 == 0:
                print(f"   Epoch {epoch} | Batch {batch_idx+1}/{len(train_loader)} | Loss: {loss.item():.4f}")

        avg_loss = total_loss / len(train_loader)

        # Validate
        model.eval()
        val_preds = []
        val_labels = []
        with torch.no_grad():
            for batch in val_loader:
                input_ids = batch["input_ids"].to(device)
                attention_mask = batch["attention_mask"].to(device)
                labels = batch["labels"].to(device)
                outputs = model(input_ids=input_ids, attention_mask=attention_mask)
                preds = torch.argmax(outputs.logits, dim=1)
                val_preds.extend(preds.cpu().tolist())
                val_labels.extend(labels.cpu().tolist())

        precision, recall, f1, _ = precision_recall_fscore_support(
            val_labels, val_preds, average="binary", zero_division=0
        )

        print(f"\n   πŸ“Š Epoch {epoch}/{EPOCHS}:")
        print(f"      Loss:      {avg_loss:.4f}")
        print(f"      Precision: {precision:.4f}")
        print(f"      Recall:    {recall:.4f}")
        print(f"      F1 Score:  {f1:.4f}")

        # Save best model
        if f1 > best_f1:
            best_f1 = f1
            BERT_WEIGHTS_DIR.mkdir(parents=True, exist_ok=True)
            model.save_pretrained(str(BERT_WEIGHTS_DIR))
            tokenizer.save_pretrained(str(BERT_WEIGHTS_DIR))
            print(f"      βœ… New best model saved to {BERT_WEIGHTS_DIR}")

    print(f"\n🎯 Best F1: {best_f1:.4f}")
    print(f"βœ… Fine-tuning complete. Weights saved to: {BERT_WEIGHTS_DIR}")
    print("=" * 60)


if __name__ == "__main__":
    main()