File size: 7,425 Bytes
3dac39e | 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 | #!/usr/bin/env python3
"""Augment training data with defanged indicator variants."""
import json
import random
import re
import sys
from pathlib import Path
random.seed(42)
INPUT = Path("data/processed/enriched_5class_train_cleaned.jsonl")
OUTPUT = Path("data/processed/defanged_augmented.jsonl")
SAMPLE_RATE = 0.30
def classify_indicator(value: str) -> str:
"""Classify an indicator value as ip, domain, url, or other."""
if "://" in value or value.startswith("hxxp"):
return "url"
# IP: all digits and dots, at least 3 dots
if all(c in "0123456789.[]() " for c in value) and value.count(".") >= 3:
return "ip"
if "." in value and not all(c in "0123456789abcdef" for c in value):
return "domain"
return "other"
def defang_ip(text: str) -> str:
"""Defang an IP address with random style."""
style = random.choice(["bracket", "bracket_space"])
if style == "bracket":
return text.replace(".", "[.]")
else:
return text.replace(".", " [ . ] ")
def defang_domain(text: str) -> str:
"""Defang a domain with random style."""
style = random.choice(["bracket", "bracket_space"])
if style == "bracket":
return text.replace(".", "[.]")
else:
return text.replace(".", "[ . ]")
def defang_url(text: str) -> str:
"""Defang a URL: protocol + domain part."""
result = text.replace("https://", "hxxps://").replace("http://", "hxxp://")
# Also defang dots in the domain portion (before first /)
proto_end = result.find("://")
if proto_end >= 0:
after_proto = proto_end + 3
slash_pos = result.find("/", after_proto)
if slash_pos == -1:
slash_pos = len(result)
domain_part = result[after_proto:slash_pos]
style = random.choice(["bracket", "bracket_space"])
if style == "bracket":
domain_defanged = domain_part.replace(".", "[.]")
else:
domain_defanged = domain_part.replace(".", "[ . ]")
result = result[:after_proto] + domain_defanged + result[slash_pos:]
return result
def defang_span_text(text: str, indicator_value: str) -> str | None:
"""Defang the text of a span. Returns new text or None if not defangable."""
itype = classify_indicator(indicator_value)
if itype == "ip":
return defang_ip(text)
elif itype == "domain":
return defang_domain(text)
elif itype == "url":
return defang_url(text)
return None
def augment_example(example: dict) -> dict | None:
"""Create a defanged copy of an example. Returns None if nothing to defang."""
text = example["text"]
spans = example["spans"]
# Collect all indicator spans with positions, sorted by start offset
indicator_spans = []
for label, positions in spans.items():
if label.startswith("Indicator:"):
indicator_value = label.split(": ", 1)[1]
for start, end in positions:
indicator_spans.append((start, end, label, indicator_value))
if not indicator_spans:
return None
# Sort by start position
indicator_spans.sort(key=lambda x: x[0])
# Try to defang each indicator span, track replacements
replacements = [] # (old_start, old_end, new_text)
for start, end, label, indicator_value in indicator_spans:
old_text = text[start:end]
new_text = defang_span_text(old_text, indicator_value)
if new_text and new_text != old_text:
replacements.append((start, end, new_text))
if not replacements:
return None
# Build new text and offset mapping
# Process replacements from end to start to not mess up offsets
# But we need a forward pass to compute cumulative offset shifts
# Compute cumulative offset adjustments
# For each position in original text, compute how much it shifts
shifts = [] # (original_pos, delta) - at original_pos, cumulative delta changes
cumulative = 0
for old_start, old_end, new_text in replacements:
old_len = old_end - old_start
new_len = len(new_text)
delta = new_len - old_len
shifts.append((old_start, old_end, delta, cumulative))
cumulative += delta
# Build new text
new_text_parts = []
prev_end = 0
for old_start, old_end, new_text in replacements:
new_text_parts.append(text[prev_end:old_start])
new_text_parts.append(new_text)
prev_end = old_end
new_text_parts.append(text[prev_end:])
new_full_text = "".join(new_text_parts)
# Adjust all span offsets
def adjust_offset(pos: int) -> int:
"""Adjust an original offset to account for replacements."""
cum = 0
for old_start, old_end, new_text in replacements:
old_len = old_end - old_start
new_len = len(new_text)
delta = new_len - old_len
if pos <= old_start:
break
elif pos >= old_end:
cum += delta
else:
# pos is inside a replacement - scale proportionally
# This handles the span endpoints that ARE the replacement
frac = (pos - old_start) / old_len
cum += int(frac * delta)
break
return pos + cum
new_spans = {}
for label, positions in spans.items():
new_positions = []
for start, end in positions:
new_start = adjust_offset(start)
new_end = adjust_offset(end)
new_positions.append([new_start, new_end])
new_spans[label] = new_positions
new_example = {
"text": new_full_text,
"spans": new_spans,
"info": {**example.get("info", {}), "source": "defanged_augment"},
}
return new_example
def main():
examples = []
with open(INPUT) as f:
for line in f:
examples.append(json.loads(line))
# Find examples with Indicator spans
indicator_examples = [
ex for ex in examples
if any(k.startswith("Indicator:") for k in ex["spans"])
]
print(f"Total examples: {len(examples)}")
print(f"Examples with Indicator spans: {len(indicator_examples)}")
# Sample 30%
sampled = random.sample(indicator_examples, int(len(indicator_examples) * SAMPLE_RATE))
print(f"Sampled for augmentation: {len(sampled)}")
augmented = []
defanged_count = 0
for ex in sampled:
result = augment_example(ex)
if result:
augmented.append(result)
# Count defanged indicators
for label in result["spans"]:
if label.startswith("Indicator:"):
defanged_count += len(result["spans"][label])
print(f"Successfully augmented: {len(augmented)}")
print(f"Total indicator spans in augmented data: {defanged_count}")
with open(OUTPUT, "w") as f:
for ex in augmented:
f.write(json.dumps(ex, ensure_ascii=False) + "\n")
print(f"Written to {OUTPUT}")
# Verify a few examples
print("\n=== Sample verification ===")
for ex in augmented[:3]:
print(f"\nText: {ex['text'][:120]}...")
for label, positions in ex["spans"].items():
if label.startswith("Indicator:"):
for s, e in positions:
print(f" {label}: '{ex['text'][s:e]}' [{s}:{e}]")
if __name__ == "__main__":
main()
|