How dpr23/anomaly_project works
This is a small cold-store IoT temperature anomaly detector. The documented workflows are training an LSTM autoencoder on synthetic normal readings with MLflow tracking, and submitting 24 readings through a Gradio interface for reconstruction-error classification. The graph separates training, serving, and the model artifact. Note: the tree contains lstm_autoencoder (1).pth, while the serving code loads lstm_autoencoder.pth; the excerpts do not establish that the listed artifact is the one the app loads.
Main entry points: train.py Training script · app.py Anomaly detector · lstm_autoencoder (1).pth Weights file
Components 9
Model training
- Training script
train.py - Normal sequences
train.py - LSTM autoencoder
train.py - MLflow tracking
Detection service
- Gradio interface
app.py - Anomaly detector
app.py - LSTM autoencoder
app.py
Model artifact
- Weights file
lstm_autoencoder (1).pth
Other components
- Operator
Connections 9
- Training script → Normal sequences: generates
train.py - Training script → LSTM autoencoder: trains
train.py - Training script → MLflow tracking: logs metrics
train.py - Training script → Weights file: saves weights
train.py - Operator → Gradio interface: enters readings
app.py - Gradio interface → Anomaly detector: submits input
app.py - Anomaly detector → LSTM autoencoder: reconstructs
app.py - Anomaly detector → Gradio interface: returns status
app.py - LSTM autoencoder → Weights file: loads weights