ferrifuzz

Fast fuzzy logic for Python, with a Rust core. Mamdani, Sugeno (TSK) and interval type-2 inference, linguistic hedges, six t-norms / t-conorms, fuzzy time series, and plotting helpers — with error messages that tell you what you got wrong.

pip install ferrifuzz                 # core, no dependencies
pip install "ferrifuzz[plot]"         # + numpy and matplotlib for ferrifuzz.plot

Prebuilt wheels for Windows, macOS and Linux, Python 3.9+. No Rust needed.

Documentation: ferrifuzz.bubblelamp.space (GitHub Pages mirror) · VS Code extension: Ferrifuzz

A system in 15 lines

import ferrifuzz as ff

service = ff.Variable("service", 0, 10)
service["poor"] = ff.mf.trapezoid(0, 0, 2, 5)
service["good"] = ff.mf.triangle(2, 5, 8)
service["excellent"] = ff.mf.trapezoid(5, 8, 10, 10)

tip = ff.Variable("tip", 0, 30).automf(["low", "medium", "high"])   # evenly spaced terms

system = ff.FuzzySystem()                  # Mamdani, min/max, centroid
system.add_input(service)
system.add_output(tip)
system += "IF service IS poor THEN tip IS low"
system += "IF service IS good THEN tip IS medium"
system += "IF service IS excellent OR service IS very good THEN tip IS high"

print(system({"service": 7.5}))            # {'tip': 20.82...}

Rule syntax

IF <condition> THEN <output> IS <term> [AND <output> IS <term> ...] [WITH <weight>]

IF temp IS hot AND humidity IS NOT low THEN fan IS fast
IF (temp IS hot OR temp IS very warm) AND wind IS calm THEN fan IS fast AND alarm IS on WITH 0.8

• Keywords IF THEN AND OR IS NOT WITH are case-insensitive; names are single words (use _).
• AND binds tighter than OR; parentheses are allowed.
• Hedges after IS [NOT]: very (μ²), extremely (μ³), somewhat / more_or_less / slightly (√μ), plus (μ^1.25), minus (μ^0.75).
• WITH w scales the rule's firing strength (0 ≤ w ≤ 1).

When something is wrong, it says so

system += "IF service IS gud THEN tip IS hihg"
system({"service": 7.5})

ferrifuzz.ModelError: invalid fuzzy system (2 problems):
  - rule #4 `IF service IS gud THEN tip IS hihg`: input 'service' has no term 'gud' (did you mean 'good'?); its terms are: 'poor', 'good', 'excellent'
  - rule #4 `IF service IS gud THEN tip IS hihg`: output 'tip' has no term 'hihg' (did you mean 'high'?); its terms are: 'low', 'medium', 'high'

system.validate() lists errors and warnings (unused inputs, duplicate or conflicting rules, terms that are zero over their whole range) without raising.

Inputs outside a variable's range are clamped to it (FuzzySystem(clamp_inputs=False) to turn that off). Every output is always returned — nan if no rule fired.

What's in the box

import ferrifuzz.plot as fp
fp.plot_variable(service)
fp.plot_output(system, {"service": 7.5}, "tip")     # aggregated set + centroid
fp.plot_response(system, "service", "tip")

Time series:

from ferrifuzz.fts import FuzzyTimeSeries

enrollments = [13055, 13563, 13867, 14696, 15460, 15311, 15603, 15861, 16807, 16919, 16388,
               15433, 15497, 15145, 15163, 15984, 16859, 18150, 18970, 19328, 19337, 18876]
model = FuzzyTimeSeries(n_partitions=7, order=1, universe=(13000, 20000)).fit(enrollments)
print(model.evaluate(enrollments))      # {'rmse': ..., 'mae': ..., 'mape': ...}
print(model.predict(enrollments[-1:], steps=2))

Editor support

Type hints and docstrings ship with the package, so hover help and autocompletion of option names (defuzz_method=" → centroid, bisector, ...) work in VS Code, PyCharm and Jupyter.

The Ferrifuzz extension (VS Code Marketplace; Open VSX for Cursor, VSCodium and Windsurf) adds, inside rule strings, highlighting, completion of variables, terms and hedges, and "did you mean" warnings with quick fixes; plus ff… snippets (ffmamdani, ffsugeno, ffit2sys, ...) and a warning when the editor's Python interpreter has no ferrifuzz.

Citing

If ferrifuzz helps your coursework, thesis or paper, please cite it:

@software{bordbar_ferrifuzz,
  author  = {Bordbar, Parsa},
  title   = {ferrifuzz: fast fuzzy logic for Python and Rust},
  version = {0.2.1},
  year    = {2026},
  url     = {https://github.com/ParsaBordbar/ferrifuzz}
}

License

Apache-2.0. If you redistribute ferrifuzz or build on it, keep the NOTICE file (attribution to Parsa Bordbar), as the license requires.