SignalFlow Ecosystem¶

SignalFlow is a modular ecosystem of Python packages for algorithmic trading. Each package focuses on a specific domain while sharing the core framework's component registry, data containers, and Flow patterns.

Packages¶

signalflow-trading v0.8.4 (Core)

Core framework: Dataset, Transform/FeaturePipe, ForecastModel, Flow, Engine, Run. Component registry, Polars-first processing, deploy-is-data YAML serialization.
```
pip install signalflow-trading
```
GitHub
signalflow-ta v0.8.2

248 technical-indicator features across 8 modules (momentum, overlap, volatility, volume, trend, statistics, performance, divergence) plus 21 signal detectors. Physics-based market analogs and AutoFeatureNormalizer.
```
pip install "signalflow-trading[ta]"
```
GitHub
signalflow-labs v0.8.2

Neural encoders (LSTM, GRU, Transformer, PatchTST, TCN, TSMixer, InceptionTime), classification heads, and an RL strategy. Built on PyTorch.
```
pip install "signalflow-trading[labs]"
```
GitHub

Architecture¶

All packages share the SignalFlow component registry via semantic decorators (@sf.detector, @sf.feature, @sf.transform, @sf.model, @sf.strategy). Components from any installed package are discoverable through the same registry:

import signalflow as sf
import signalflow.ta  # noqa: F401 - registers the ta components

rsi_cls = sf.registry.get(sf.ComponentType.TRANSFORM, "momentum/rsi")
sf.registry.snapshot()  # {type: [names]} across every installed package

Installing a plugin auto-registers its components via the signalflow.components entry point - no imports or wiring needed.

Dependency Chain¶

signalflow-trading              # Core (required)
├── signalflow-ta               # 248 features + 21 detectors
├── signalflow-labs             # Neural encoders, RL strategy
└── sf-custom                   # User components (entry-point autodiscovery)

Extension packages use Python namespace packages under signalflow.*. Custom packages register through the signalflow.components entry point.