LSTM Autoencoder + Graph Neural Network for IoT sensor anomaly detection with OTA deployment.

• Pure NumPy LSTM Autoencoder — no PyTorch or TensorFlow required
• Sensor Graph — adjacency matrix + graph convolution for spatial relationships
• Anomaly Detector — auto-calibrating threshold from normal data
• Synthetic Data Generator — sinusoidal sensor patterns with spike/drift/dropout anomalies
• OTA Deployer — model versioning with numpy .npz format
• Docker-ready

pip install -e .
python -m iot_anomaly

src/iot_anomaly/
  data_generator.py   # SyntheticSensorData
  lstm_autoencoder.py # Pure numpy LSTM autoencoder
  graph.py            # SensorGraph with adjacency matrix
  detector.py         # AnomalyDetector
  ota.py              # OTADeployer
tests/
  test_data_generator.py
  test_lstm_autoencoder.py
  test_graph.py
  test_detector.py
  test_ota.py

from iot_anomaly.data_generator import SyntheticSensorData
from iot_anomaly.detector import AnomalyDetector

gen = SyntheticSensorData(n_sensors=5, seq_len=50)
X_normal = gen.generate_normal(n_samples=100)
X_anomalous = gen.generate_anomalous(n_samples=20, anomaly_type="spike")

detector = AnomalyDetector(n_sensors=5, hidden_dim=16)
detector.fit(X_normal)
predictions = detector.predict(X_anomalous)

pytest tests/ -v