tidystl 0.2.0

STL robustness evaluation with explicit semantics.

tidystl

STL robustness evaluation for Python. Explicit semantics, batch-first, extensible backends.

import numpy as np
from tidystl import Signal, parse, robustness

phi = parse("G[0,5](F[0,2](velocity >= 10))")

velocity_batch = np.full((100, 200), 12.0)  # N=100 traces, T=200 samples

sig = Signal.from_dict(
    times=np.linspace(0, 10, 200),
    values={"velocity": velocity_batch},  # shape: (N, T)
)

rho = robustness(phi, sig)  # shape: (N, T)

Features

• Minimal dependencies: numpy and lark only; no C++ build, no MATLAB
• Explicit semantics: boundary handling, interpolation, and end-of-trace behavior are documented and switchable per backend
• Cross-validated: regression suites against reference tools (Breach, RTAMT, and others) using their ground-truth data
• Batch-first: native (N, S, T) tensor layout
• Backend-centric: swap evaluation strategy without changing formula or signal
• Full STL: arbitrary nesting including the Until operator

Use cases

tidySTL is useful when you want to evaluate STL robustness in a small, explicit, and testable way.

• Evaluate STL robustness Compute robustness values of STL formulas over real-valued signals.

• Test STL semantics Check boundary-sensitive behavior such as interpolation, sampling, interval endpoints, and temporal aggregation, and compare the results with other STL tools when needed.

• Experiment with semantics Implement alternative robustness semantics by changing only the relevant evaluation components.

• Build research prototypes Use tidySTL as a lightweight STL evaluation layer inside monitoring, falsification, learning, or control workflows.

Installation

pip install tidystl

Requires Python 3.11+, numpy, lark.

The compatibility backends (Breach, RTAMT, STLCG++, TaLiRo, py-MTL, and more) live in a separate package; install it and register it once:

pip install tidystl-compat

import tidystl
import tidystl_compat
from tidystl import list_backends

tidystl.use(tidystl_compat)

print(list_backends())  # the full set registered, e.g. ['breach', 'native', ...]
# every listed name is usable as backend="...", e.g. backend="breach"

Only native is built into tidystl itself; list_backends() always reflects what is currently registered.

Documentation

For the full user manual (installation, signals, specification language, backend semantics, and worked examples), see docs/usage.md; for grammar tables, call signatures, and API contracts, see docs/reference.md.

License

MIT License. Copyright (c) 2026 National Institute of Advanced Industrial Science and Technology (AIST). Developed by Sota Sato.