File size: 14,374 Bytes
e520bf1 | 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 | """
feature_engineering.py
======================
Feature pipeline for the CYB009 baseline classifier.
Predicts `vulnerability_class` (8-class vulnerability classification)
from per-vulnerability features on the CYB009 sample dataset.
CSV inputs:
vuln_summary.csv (primary, one row per vulnerability,
2,638 vulnerabilities)
asset_inventory.csv (per-asset registry, joined for asset
context features)
vulnerability_records.csv (per-timestep trajectory; reserved)
vuln_lifecycle_events.csv (discrete event log; reserved)
Target classes (8):
auth_access_control, cryptographic_failure, information_disclosure,
injection_family, logic_flaw, memory_corruption, misconfiguration,
supply_chain_weakness
Why this task (and why not the more obvious targets)
----------------------------------------------------
The CYB009 README lists 11 suggested use cases. We piloted every
README-headline target on the sample dataset and found the sample
has pervasive structural leakage that makes most targets either
trivially solvable via oracle features or unlearnable after honest
leakage removal:
- `exploit_maturity_final` (4-class) is structurally leaky via
`cvss_temporal_score_final`: CVSS v3.1 computes temporal score from
base score using Exploit Code Maturity multipliers (0.91 / 0.94 /
0.97 / 1.00 for unproven / PoC / functional / weaponised), so the
cvss_temporal/cvss_base ratio clusters near-deterministically per
maturity tier (0.80 / 0.83 / 0.85 / 0.88 in the data). Drop
cvss_temporal -> accuracy collapses from 0.74 to 0.31 (below
majority 0.36).
- `remediation_status` / `patch_status` / `lifecycle_phase`
(per-timestep) form a tightly-coupled state machine. lifecycle_phase
= `residual_risk_review` -> 100% `remediated`. `patch_status =
deployed` -> 100% `remediated`. Any two of the three deterministically
pin the third.
- `severity_class` is 100% derived from `cvss_base_score` via CVSS
v3.1 boundaries (low=0.1-3.9, medium=4.0-6.9, high=7.0-8.9,
critical=9.0-10.0). Trivial if cvss_base included; below majority
(acc 0.55 vs majority 0.51) without it.
- All seven binary flags (`exploitation_occurred_flag`, `zero_day_flag`,
`cisa_kev_flag`, `supply_chain_propagation_flag`,
`remediation_success_flag`, `sla_compliance_flag`,
`false_positive_flag`) are at-or-below majority after honest
leakage removal of the event-time sentinels
(`time_to_exploit_days`, `time_to_remediate_days`, `patch_lag_days`,
`risk_score_composite`). See leakage_diagnostic.json.
`vulnerability_class` is the only README-suggested target that learns
honestly on the sample: acc 0.24, macro-F1 0.22, ROC-AUC 0.69 vs
majority baseline 0.18. Modest +6pp lift over majority - the weakest
baseline in the XpertSystems CYB catalog by design. The full ~487k-row
product would tighten per-class signal materially.
The model card frames this honestly: the strongest finding on CYB009
is the comprehensive leakage diagnostic rather than the modest
classifier performance. Buyers planning CYB009 ML work should read
the diagnostic first.
Leakage audit
-------------
Excluded as outcome leaks for this target:
1. `exploit_maturity_final` - the target's natural pair via the CVSS
v3.1 temporal-score machinery.
2. Event-time sentinel oracles dropped as precaution (not directly
leaky for vulnerability_class but indirectly via flag fields):
`time_to_exploit_days`, `time_to_remediate_days`, `patch_lag_days`,
`risk_score_composite`.
3. `cvss_temporal_score_final` excluded because of the CVSS v3.1
maturity-multiplier structural encoding.
`severity_class` is KEPT as a one-hot feature because it's a derived
view of `cvss_base_score` rather than the target.
Binary post-hoc flags are KEPT as legitimate observables that a SOC
analyst would have at decision time. They contribute modest real
signal (a few pp accuracy).
Public API
----------
build_features(vuln_summary_path, asset_inventory_path)
-> (X, y, ids, meta)
transform_single(record, meta, asset_lookup=None) -> np.ndarray
save_meta(meta, path) / load_meta(path)
build_asset_lookup(asset_inventory_path) -> dict
License
-------
Ships with the public model on Hugging Face under CC-BY-NC-4.0,
matching the dataset license. See README.md.
"""
from __future__ import annotations
import json
from pathlib import Path
from typing import Any
import numpy as np
import pandas as pd
# ---------------------------------------------------------------------------
# Label space
# ---------------------------------------------------------------------------
# Eight vulnerability classes from the CYB009 sample. The README claims
# 10 classes but only 8 exist in the sample data.
LABEL_ORDER = [
"auth_access_control",
"cryptographic_failure",
"information_disclosure",
"injection_family",
"logic_flaw",
"memory_corruption",
"misconfiguration",
"supply_chain_weakness",
]
LABEL_TO_INT = {lbl: i for i, lbl in enumerate(LABEL_ORDER)}
INT_TO_LABEL = {i: lbl for lbl, i in LABEL_TO_INT.items()}
# ---------------------------------------------------------------------------
# Identifier and target columns
# ---------------------------------------------------------------------------
ID_COLUMNS = ["vuln_id", "asset_id", "org_id"]
TARGET_COLUMN = "vulnerability_class"
# Outcome-leak columns excluded from features.
EXCLUDED_FROM_FEATURES = [
"time_to_exploit_days", # -1 sentinel oracle
"time_to_remediate_days", # 120 sentinel oracle
"patch_lag_days", # likely similar sentinel
"risk_score_composite", # computed from flag fields
"exploit_maturity_final", # indirect leak via CVSS temporal
"cvss_temporal_score_final", # near-deterministic per maturity tier
]
# ---------------------------------------------------------------------------
# Per-vulnerability numeric features
# ---------------------------------------------------------------------------
VULN_NUMERIC_FEATURES = [
"cvss_base_score",
"epss_score_final",
"exploitation_occurred_flag",
"zero_day_flag",
"cisa_kev_flag",
"supply_chain_propagation_flag",
"compensating_control_flag",
"false_positive_flag",
"remediation_success_flag",
"sla_compliance_flag",
]
VULN_CATEGORICAL_FEATURES = [
"severity_class", # 4 values; CVSS-derived but useful as feature
]
# ---------------------------------------------------------------------------
# Asset features (joined on asset_id from asset_inventory.csv)
# ---------------------------------------------------------------------------
ASSET_NUMERIC_FEATURES = [
"scanner_coverage",
"patch_mgmt_maturity",
"mean_time_to_remediate_days",
"sla_critical_days",
"sla_high_days",
"sla_medium_days",
"internet_exposed_flag",
"sbom_depth_score",
]
ASSET_CATEGORICAL_FEATURES = [
"asset_type", # 12 values
"criticality_tier", # 4 values
"environment_type", # 8 values
"os_family", # 6 values
]
# ---------------------------------------------------------------------------
# Engineered features
# ---------------------------------------------------------------------------
def _add_engineered_features(df: pd.DataFrame) -> pd.DataFrame:
"""
Five engineered features for vulnerability_class discrimination.
Note: no temporal-CVSS-derived features (those leak via the CVSS
v3.1 exploit-code-maturity machinery).
"""
df = df.copy()
# 1. Log-scaled EPSS. EPSS is heavy-tailed.
df["log_epss"] = np.log1p(
df["epss_score_final"].clip(lower=0)
).astype(float)
# 2. High-CVSS indicator. CVSS >= 7.0 (high or critical).
df["is_high_cvss"] = (df["cvss_base_score"] >= 7.0).astype(int)
# 3. Exposure x severity composite. Internet-exposed high-severity
# vulns are often weighted differently per class.
df["exposure_severity_composite"] = (
df.get("internet_exposed_flag", 0) * df["cvss_base_score"]
).astype(float)
# 4. Flag count: total number of risk flags raised. Different vuln
# classes have different baseline flag patterns.
flag_cols = [
"exploitation_occurred_flag", "zero_day_flag", "cisa_kev_flag",
"supply_chain_propagation_flag", "compensating_control_flag",
"false_positive_flag",
]
df["risk_flag_count"] = sum(df.get(c, 0) for c in flag_cols)
# 5. EPSS x CVSS composite.
df["epss_x_base"] = (
df["epss_score_final"] * df["cvss_base_score"]
).astype(float)
return df
# ---------------------------------------------------------------------------
# Public API
# ---------------------------------------------------------------------------
def build_features(
vuln_summary_path: str | Path,
asset_inventory_path: str | Path,
) -> tuple[pd.DataFrame, pd.Series, pd.Series, dict[str, Any]]:
"""
Load vuln_summary.csv, join asset_inventory.csv, drop target +
identifiers + outcome leaks, engineer features, one-hot encode,
return (X, y, ids, meta).
"""
vulns = pd.read_csv(vuln_summary_path)
assets = pd.read_csv(asset_inventory_path)
y = vulns[TARGET_COLUMN].map(LABEL_TO_INT)
if y.isna().any():
bad = vulns.loc[y.isna(), TARGET_COLUMN].unique()
raise ValueError(f"Unknown vulnerability_class values: {bad}")
y = y.astype(int)
ids = vulns["vuln_id"].copy()
asset_cols_needed = (
["asset_id"] + ASSET_NUMERIC_FEATURES + ASSET_CATEGORICAL_FEATURES
)
vulns = vulns.merge(
assets[asset_cols_needed], on="asset_id", how="left",
)
vulns = vulns.drop(
columns=ID_COLUMNS + [TARGET_COLUMN] + EXCLUDED_FROM_FEATURES,
errors="ignore",
)
vulns = _add_engineered_features(vulns)
numeric_features = (
VULN_NUMERIC_FEATURES
+ ASSET_NUMERIC_FEATURES
+ [
"log_epss", "is_high_cvss", "exposure_severity_composite",
"risk_flag_count", "epss_x_base",
]
)
numeric_features = [c for c in numeric_features if c in vulns.columns]
X_numeric = vulns[numeric_features].astype(float)
all_categorical = VULN_CATEGORICAL_FEATURES + ASSET_CATEGORICAL_FEATURES
categorical_levels: dict[str, list[str]] = {}
blocks: list[pd.DataFrame] = []
for col in all_categorical:
if col not in vulns.columns:
continue
levels = sorted(vulns[col].dropna().unique().tolist())
categorical_levels[col] = levels
block = pd.get_dummies(
vulns[col].astype("category").cat.set_categories(levels),
prefix=col, dummy_na=False,
).astype(int)
blocks.append(block)
X = pd.concat(
[X_numeric.reset_index(drop=True)]
+ [b.reset_index(drop=True) for b in blocks],
axis=1,
).fillna(0.0)
meta = {
"feature_names": X.columns.tolist(),
"numeric_features": numeric_features,
"categorical_levels": categorical_levels,
"label_to_int": LABEL_TO_INT,
"int_to_label": INT_TO_LABEL,
"outcome_leak_excluded": EXCLUDED_FROM_FEATURES,
}
return X, y, ids, meta
def transform_single(
record: dict | pd.DataFrame,
meta: dict[str, Any],
asset_lookup: dict | None = None,
) -> np.ndarray:
"""Encode a single vulnerability record for inference."""
if isinstance(record, dict):
df = pd.DataFrame([record.copy()])
else:
df = record.copy()
if asset_lookup is not None and "asset_id" in df.columns:
asset_id = df["asset_id"].iloc[0]
asset_feats = asset_lookup.get(asset_id, {})
for k, v in asset_feats.items():
if k not in df.columns:
df[k] = v
df = _add_engineered_features(df)
numeric = pd.DataFrame({
col: df.get(col, pd.Series([0.0] * len(df))).astype(float).values
for col in meta["numeric_features"]
})
blocks: list[pd.DataFrame] = [numeric]
for col, levels in meta["categorical_levels"].items():
val = df.get(col, pd.Series([None] * len(df)))
block = pd.get_dummies(
val.astype("category").cat.set_categories(levels),
prefix=col, dummy_na=False,
).astype(int)
for lvl in levels:
cname = f"{col}_{lvl}"
if cname not in block.columns:
block[cname] = 0
block = block[[f"{col}_{lvl}" for lvl in levels]]
blocks.append(block)
X = pd.concat(blocks, axis=1).fillna(0.0)
X = X.reindex(columns=meta["feature_names"], fill_value=0.0)
return X.values.astype(np.float32)
def save_meta(meta: dict[str, Any], path: str | Path) -> None:
serializable = {
"feature_names": meta["feature_names"],
"numeric_features": meta["numeric_features"],
"categorical_levels": meta["categorical_levels"],
"label_to_int": meta["label_to_int"],
"int_to_label": {str(k): v for k, v in meta["int_to_label"].items()},
"outcome_leak_excluded": meta.get("outcome_leak_excluded", []),
}
with open(path, "w") as f:
json.dump(serializable, f, indent=2)
def load_meta(path: str | Path) -> dict[str, Any]:
with open(path) as f:
meta = json.load(f)
meta["int_to_label"] = {int(k): v for k, v in meta["int_to_label"].items()}
return meta
def build_asset_lookup(asset_inventory_path: str | Path) -> dict[str, dict]:
"""Build {asset_id: {asset feature values}} for inference-time lookup."""
assets = pd.read_csv(asset_inventory_path)
cols = ASSET_NUMERIC_FEATURES + ASSET_CATEGORICAL_FEATURES
out = {}
for _, row in assets.iterrows():
out[row["asset_id"]] = {c: row[c] for c in cols if c in assets.columns}
return out
if __name__ == "__main__":
import sys
base = Path(sys.argv[1]) if len(sys.argv) > 1 else Path("/mnt/user-data/uploads")
X, y, ids, meta = build_features(
base / "vuln_summary.csv",
base / "asset_inventory.csv",
)
print(f"X shape: {X.shape}")
print(f"y shape: {y.shape}")
print(f"n_features: {len(meta['feature_names'])}")
print(f"label distribution:\n{y.map(INT_TO_LABEL).value_counts()}")
print(f"X has NaN: {X.isnull().any().any()}")
|