AnomalyDetect

High-Performance Anomaly Detection Framework

---

AnomalyDetect is a production-ready anomaly detection framework built for speed and scale. It combines parallel model training, multi-architecture ensembles, and GPU acceleration into a single, easy-to-use package.

Why AnomalyDetect?

Building robust anomaly detectors is hard. Building fast ones shouldn't be.

• 5 Models, 1 Click — Train LSTM, CNN, Dense, Transformer, and GRU simultaneously
• GPU-Native — Full CUDA and DirectML support out of the box
• Intelligent Fallback — Automatically falls back to CPU if GPU training fails
• Production-Ready — REST API + Web UI included
• Open Source — MIT licensed, community driven

Quick Start

# Install
pip install -r requirements.txt

# Run the web interface
python server.py

Open http://localhost:8000 in your browser.

Or use the CLI:

python main.py --epochs 10 --batch_size 32 --n_models 5 --device dml

Python API

from parallel_engine import ParallelTrainer
from ensemble import EnsemblePredictor
from data_loader import get_dataloaders

# Load your data
train_loader, test_loader = get_dataloaders("synthetic", batch_size=32, seq_len=64)

# Train 5 models in parallel
trainer = ParallelTrainer(n_models=5, device="dml")
results = trainer.train(model_configs, train_loader, val_loader)

# Ensemble predictions
ensemble = EnsemblePredictor(results)
metrics = ensemble.evaluate_ensemble(test_loader, device="dml")
print(f"AUC-ROC: {metrics['auc_roc']:.4f}")

Features

Feature	Description
Parallel Training	Train multiple models simultaneously using multiprocessing
Multi-Architecture Ensemble	LSTM, CNN, Dense, Transformer, GRU vote together
GPU Acceleration	Native CUDA (NVIDIA) and DirectML (AMD) support
Auto-Fallback	Seamlessly falls back to CPU on GPU errors
Real-Time Dashboard	Live training progress and metrics via React UI
REST API	Programmatic access for integration into your apps

Supported Hardware

Device	Use Case
CPU	Compatible everywhere
CUDA	NVIDIA GPUs
DirectML	AMD GPUs (Windows)

Architecture

AnomalyDetect/
├── main.py              # CLI entry point
├── server.py           # FastAPI server + Web UI
├── parallel_engine.py   # Distributed training
├── ensemble.py        # Ensemble voting
├── models.py          # Model zoo (LSTM, CNN, Dense, Transformer, GRU)
├── data_loader.py    # Data pipelines
├── trainer.py        # Training utilities
├── benchmark.py      # Performance benchmarking
├── config.json      # Configuration
├── frontend/        # React dashboard
└── requirements.txt

Configuration

Customize behavior in config.json:

{
  "system": {
    "use_gpu": true,
    "use_dml": true,
    "num_workers_ratio": 0.75
  },
  "training": {
    "default_epochs": 10,
    "batch_size": 32,
    "learning_rate": 0.001
  }
}

REST API

Endpoint	Method	Description
/config	GET	Current configuration
/config	POST	Update configuration
/train	POST	Start training job
/results	GET	Training results
/logs	GET	Live training logs (SSE)
/stop	POST	Stop active training

Requirements

• Python 3.11+
• PyTorch 2.0+
• (Optional) torch-directml for AMD GPU support

License

MIT License — see LICENSE for details.

---

AnomalyDetect — Find anomalies before they find you.