---

Install

pip install vigil-drift

---

The Problem

In production ML systems, the data your model was trained on eventually stops looking like the data it receives — concept drift. For network security, attacks evolve. Most drift detectors require labels (unavailable in real-time) or only tell you that drift happened, not what drifted.

Vigil solves both.

---

What It Does

Capability	How
🔍 Detect concept drift	Replicated T-Test (r=15) on autoencoder reconstruction errors — no labels needed
🆕 Identify novel classes	KDE density estimation on a frozen mirror autoencoder (A_KC)
📊 Explain the drift	DriftAttributor ranks input features by reconstruction error delta
⚡ Serve at scale	FastAPI REST service with /fit and /detect endpoints
📡 Stream-native	Kafka producer/consumer pipeline for live network traffic
📈 Track experiments	MLflow logs every chunk's drift severity, novelty, and attribution
✈️ Auto-retrain	Airflow DAGs trigger retraining with quality gate when drift accumulates
🖥️ Visualize live	SOC-style Streamlit dashboard with real-time charts

---

Architecture

                                    ┌─────────────────────────────────────────────┐
                                    │                   Vigil                     │
                                    │                                             │
 Data Stream                        │  ┌─────────────┐   ┌─────────────────────┐ │
 CSV · Kafka · REST ──► Chunk  ─────┼─►│ Autoencoder │──►│ Reconstruction      │ │
                       200 rows     │  │  A (adapts) │   │ Errors A            │ │   ┌──────────┐
                           │        │  └─────────────┘   └──────────┬──────────┘ │──►│  MLflow  │
                           │        │                               │             │   └──────────┘
                           │        │                    ┌──────────▼──────────┐  │
                           │        │                    │  Replicated T-Test  │  │   ┌──────────┐
                           │        │                    │  r=15   α=0.05      ├──┼──►│ FastAPI  │
                           │        │                    └──────────┬──────────┘  │   └──────────┘
                           │        │                               │ Drift?      │
                           │        │                    ┌──────────▼──────────┐  │   ┌──────────┐
                           │        │                    │  DriftAttributor    ├──┼──►│Streamlit │
                           │        │                    │  ★ novel contrib    │  │   └──────────┘
                           │        │                    │  ranks Δerror/feat  │  │
                           │        │                    └─────────────────────┘  │   ┌──────────┐
                           │        │                                             │──►│ Airflow  │
                           │        │  ┌─────────────┐   ┌─────────────────────┐ │   │ DAG      │
                           └────────┼─►│ Autoencoder │──►│ Reconstruction      │ │   └──────────┘
                                    │  │ A_KC(frozen)│   │ Errors A_KC         │ │
                                    │  └─────────────┘   └──────────┬──────────┘ │
                                    │                    ┌──────────▼──────────┐  │
                                    │                    │  KDE Novelty        │  │
                                    │                    │  Detector           │  │
                                    │                    └─────────────────────┘  │
                                    └─────────────────────────────────────────────┘

Key design decisions (arXiv:2605.29834):

• 1-D reconstruction error proxy → memory O(buffer_size) not O(n × d)
• Dual autoencoders: A adapts on drift; A_KC stays frozen → separates drift from novelty
• Replicated T-tests (r=15) reduce variance vs a single test

Novel contribution (not in the paper):

• DriftAttributor — per-feature Δerror → ranks features by drift contribution → makes alerts actionable

---

Quick Start

from vigil import Vigil

v = Vigil(feature_names=feature_cols, top_k_features=5)

# Phase 1: train on baseline traffic (offline)
v.fit(baseline_data)

# Phase 2: detect on every incoming chunk
for chunk in stream:
    result = v.detect(chunk)

    if result.drift_detected:
        print(f"⚠  severity={result.drift_severity:.2f}")
        for feat in result.attribution.top_features:
            print(f"   {feat['feature_name']}: {feat['contribution']:.1%}")

# ⚠  severity=1.00
# → service_eco_i               15.3%   (port scan signature)
# → dst_host_same_src_port_rate  12.3%  (scanning pattern)
# → srv_diff_host_rate           11.1%  (lateral movement)

---

Live Dashboard

pip install "vigil-drift[dashboard]"
streamlit run dashboard/app.py

---

REST API

pip install "vigil-drift[api]"
uvicorn api.app:app --host 0.0.0.0 --port 8000 --reload
# Swagger UI → http://localhost:8000/docs

---

Kafka Streaming

docker compose up kafka zookeeper -d
python kafka_pipeline/producer.py   # publish NSL-KDD chunks
python kafka_pipeline/consumer.py   # detect drift in real-time

[Consumer] Chunk 11 [ipsweep] ⚠ DRIFT  | severity=1.00
             → service_eco_i(15.3%), dst_host_same_src_port_rate(12.3%)
[Consumer] Chunk 24 [normal] ⚠ DRIFT   | severity=0.33
             → root_shell(27.7%), service_telnet(21.5%)

---

Tests

pytest tests/ -v --cov=vigil
# 14 passed, 81% coverage

---

Research Basis

Implements and extends:

"Open World Autoencoding Drift Detection with Novel Class Recognition in Tabular Non-stationary Data Streams" arXiv:2605.29834

---

Links

• GitHub: https://github.com/Venkateswara-Sahu/OWADD
• Website: https://venkateswara-sahu.github.io/OWADD/
• PyPI: https://pypi.org/project/vigil-drift/

---

MIT © Venkateswara Sahu