PySTL

A unified API for Signal Temporal Logic (STL) robustness evaluation across multiple semantics and computational backends.

PySTL consolidates five STL quantitative semantics — Classical, Smooth, Cumulative, AGM, and D-GMSR — under one consistent interface. Switch semantics or backends without rewriting your formulas.

from pystl import Predicate, Interval, create_semantics

phi = Predicate("x_positive", fn=lambda s, t: s[t, 0]).always(Interval(0, 4))

rho_classical = phi.evaluate(signal, create_semantics("classical", backend="numpy"))
rho_smooth    = phi.evaluate(signal, create_semantics("smooth",    backend="jax"))
rho_agm       = phi.evaluate(signal, create_semantics("agm",       backend="torch"))

Contents

• Installation
• Quick Start
• A Unified API
• Backends
  • JAX
  • PyTorch
• API Reference

Supported Semantics

Semantic	Smooth	Sign-preserving	Key idea
classical	No	Yes	min/max worst-case robustness
smooth	Yes	Approx.	Softmin/softmax via logsumexp
cumulative	Piecewise	No	Integrates robustness over time
agm	Piecewise	Yes	Arithmetic-Geometric Mean blending
dgmsr	Mostly yes	Yes	Generalized means, smooth + sign-safe

Each semantic is available on three backends: numpy, jax, and torch.