File size: 8,184 Bytes
955a66f | 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 | from __future__ import annotations
from typing import Any
import numpy as np
import pandas as pd
from src.domain import (
BURST_WINDOW_MINUTES,
FEATURE_COLUMNS,
SENSITIVE_TYPES,
Split,
TEST_START_ISO,
)
from src.mapper import normalize_event
from src.parsers import parse_timestamp
from src.ruleset import maintenance_match, severity_for
def events_frame(rows: list[dict[str, Any]] | pd.DataFrame) -> pd.DataFrame:
if isinstance(rows, pd.DataFrame):
records = rows.to_dict(orient="records")
else:
records = list(rows)
cleaned = [normalize_event(row) for row in records]
frame = pd.DataFrame(cleaned)
if frame.empty:
return frame
frame["ts"] = [parse_timestamp(value) for value in frame["timestamp"]]
valid = frame["ts"].notna() & frame["event_name"].astype(str).ne("")
frame = frame.loc[valid].copy().reset_index(drop=True)
if frame.empty:
return frame
test_start = pd.Timestamp(TEST_START_ISO)
existing = frame["split"].replace("", Split.DEVELOPMENT.value)
frame["split"] = np.where(pd.to_datetime(frame["ts"], utc=True) >= test_start, Split.TEST.value, existing)
frame["hour"] = pd.to_datetime(frame["ts"], utc=True).dt.hour
frame["weekday"] = pd.to_datetime(frame["ts"], utc=True).dt.weekday
frame["weekend"] = frame["weekday"].isin([5, 6]).astype(float)
human = frame["actor_kind"].eq("human")
unusual = human & ((frame["hour"] < 6) | (frame["hour"] >= 22) | frame["weekday"].eq(6))
frame["unusual_hour"] = unusual.astype(float)
frame["in_window"] = frame["weekday"].eq(6) & frame["hour"].between(2, 5)
flags = frame.apply(
lambda row: {
"cidr_open": bool(row.get("cidr_open")),
"policy_public": bool(row.get("policy_public")),
"admin_policy": bool(row.get("admin_policy")),
"logging_off": bool(row.get("logging_off")),
"detector_off": bool(row.get("detector_off")),
"wildcard_trust": bool(row.get("wildcard_trust")),
"is_root": row.get("actor_kind") == "root",
"hardening": bool(row.get("hardening")),
},
axis=1,
)
frame["severity"] = [
severity_for(str(row.change_type), flags.iloc[idx]).value for idx, row in enumerate(frame.itertuples(index=False))
]
frame["in_maintenance"] = [
maintenance_match(
str(row.account_alias),
str(row.actor_alias),
str(row.change_type),
flags.iloc[idx],
str(row.change_ticket or ""),
bool(row.in_window),
)
for idx, row in enumerate(frame.itertuples(index=False))
]
frame["approval_status"] = np.where(frame["in_maintenance"], "approved_window", "none")
frame["approval_status"] = np.where(
frame["change_ticket"].astype(str).ne("") & ~frame["in_maintenance"],
"unknown",
frame["approval_status"],
)
frame = _pair_before_after(frame)
frame = _burst_counts(frame)
frame["is_root"] = frame["actor_kind"].eq("root").astype(float)
frame["is_service"] = frame["actor_kind"].eq("service").astype(float)
frame["is_human"] = frame["actor_kind"].eq("human").astype(float)
frame["error_event"] = frame["error_code"].astype(str).ne("").astype(float)
frame["sensitive"] = frame["change_type"].isin(SENSITIVE_TYPES) & ~frame["hardening"].astype(bool)
return frame.sort_values(["ts", "event_id"]).reset_index(drop=True)
def _pair_before_after(frame: pd.DataFrame) -> pd.DataFrame:
work = frame.sort_values("ts").copy()
before_ids: list[str] = []
before_known: list[bool] = []
after_known: list[bool] = []
for idx, row in work.iterrows():
if row["change_type"] not in SENSITIVE_TYPES or row["hardening"]:
before_ids.append("")
before_known.append(False)
after_known.append(False)
continue
prior = work.loc[
(work["ts"] < row["ts"])
& (work["account_alias"] == row["account_alias"])
& (work["resource_alias"] == row["resource_alias"])
& (work["resource_alias"].astype(str).ne("res_unknown"))
]
if prior.empty:
before_ids.append("")
before_known.append(False)
after_known.append(not bool(row.get("incomplete")))
continue
before_ids.append(str(prior.iloc[-1]["event_id"]))
before_known.append(True)
after_known.append(not bool(row.get("incomplete")))
work["before_event_id"] = before_ids
work["after_event_id"] = np.where(work["event_id"].astype(str).ne(""), work["event_id"], "")
work["has_before"] = np.array(before_known, dtype=float)
work["has_after"] = np.array(after_known, dtype=float)
work["before_state"] = np.where(work["has_before"] >= 1, "known", "unknown")
work["after_state"] = np.where(work["has_after"] >= 1, "known", "unknown")
return work
def _burst_counts(frame: pd.DataFrame) -> pd.DataFrame:
work = frame.sort_values("ts").copy()
counts = []
window = pd.Timedelta(minutes=BURST_WINDOW_MINUTES)
sensitive = work["change_type"].isin(SENSITIVE_TYPES) & ~work["hardening"].astype(bool)
for _, row in work.iterrows():
if not bool(sensitive.loc[row.name] if row.name in sensitive.index else False):
counts.append(0)
continue
mask = (
(work["actor_alias"] == row["actor_alias"])
& (work["ts"] <= row["ts"])
& (work["ts"] >= row["ts"] - window)
& sensitive
)
counts.append(int(mask.sum()))
work["burst_count"] = counts
return work
def feature_matrix(frame: pd.DataFrame) -> pd.DataFrame:
if frame.empty:
return pd.DataFrame(columns=list(FEATURE_COLUMNS))
data = {}
for column in FEATURE_COLUMNS:
if column == "severity_weight":
data[column] = frame["severity"].map({"high": 4.0, "medium": 3.2, "low": 1.8}).fillna(0.0)
elif column in frame.columns:
data[column] = pd.to_numeric(frame[column], errors="coerce").fillna(0.0)
else:
data[column] = 0.0
return pd.DataFrame(data, index=frame.index)
def attach_gold(frame: pd.DataFrame, gold: pd.DataFrame | list[dict[str, Any]] | None) -> pd.DataFrame:
out = frame.copy()
out["gold_event_ids"] = ""
out["gold_family"] = ""
out["detect_delay_min"] = np.nan
if gold is None:
return out
gold_frame = gold if isinstance(gold, pd.DataFrame) else pd.DataFrame(gold)
if gold_frame.empty:
return out
gold_frame = gold_frame.copy()
gold_frame["start_ts"] = [parse_timestamp(value) for value in gold_frame["start"]]
gold_frame["end_ts"] = [parse_timestamp(value) for value in gold_frame["end"]]
for event in gold_frame.itertuples(index=False):
lab_id = str(event.event_id)
direct = out["lab_event_id"].astype(str).eq(lab_id) | out["event_id"].astype(str).eq(lab_id)
if event.start_ts and event.end_ts:
overlap = (out["ts"] >= event.start_ts) & (out["ts"] <= event.end_ts)
if getattr(event, "account_alias", ""):
overlap = overlap & out["account_alias"].astype(str).eq(str(event.account_alias))
if getattr(event, "actor_alias", ""):
overlap = overlap & out["actor_alias"].astype(str).eq(str(event.actor_alias))
else:
overlap = pd.Series(False, index=out.index)
hit = direct | overlap
if not hit.any():
continue
current = out.loc[hit, "gold_event_ids"].astype(str)
out.loc[hit, "gold_event_ids"] = np.where(current.eq("") | current.eq("nan"), lab_id, current + "," + lab_id)
out.loc[hit, "gold_family"] = str(event.family)
delay = (pd.to_datetime(out.loc[hit, "ts"], utc=True) - pd.Timestamp(event.start_ts)).dt.total_seconds() / 60.0
out.loc[hit, "detect_delay_min"] = delay
return out
def change_view(frame: pd.DataFrame) -> pd.DataFrame:
if frame.empty:
return frame
return frame.loc[frame["change_type"].isin(SENSITIVE_TYPES)].copy()
|