LFA Toolbox

LFA Toolbox is a set of tools to create, view and play with fuzzy systems. Ideal for students to play with fuzzy systems.

For example a simple example of a fuzzy system to determine if there will be a lot of tourists given the temperature and the amount of sunshine.

• IF (temperature is hot OR sunshine is sunny), THEN (tourists is high)
• IF (temperature is warm AND sunshine is part_sunny), THEN (tourists is medium)
• IF (temperature is cold OR sunshine is cloudy), THEN (tourists is low)

Installation

The detailed install instructions can be found here docs/INSTALL.md but you basically need to do:

pip install lfa-toolbox

Features

• Singleton and Mamdani fuzzy system types
• Commonly used membership functions (Trap. MF, Triangle MF, Free shape MF,..)
• Multiple consequents
• Default rule
• NOT conditions
• Viewers (membership function viewer, linguistic variable viewer,...)

Examples

Membership functions

TODO hand-crafted mf + triangular mf + linPWMF + screenshot viewer

from matplotlib import pyplot as plt

from lfa_toolbox.core.mf.triangular_mf import TriangularMF
from lfa_toolbox.view.mf_viewer import MembershipFunctionViewer

# Create a matplotlib plot
fig, ax = plt.subplots()

# Create a triangular membership function
temp_mf = TriangularMF(-20, 25, 40)

# You can fuzzify an input value
fuzzified_value = temp_mf.fuzzify(22.5)
print("MF has been fuzzified to {:.3f}".format(fuzzified_value))

# Or you can visualize the MF using matplotlib
mfv = MembershipFunctionViewer(temp_mf, ax=ax,
                               label="Temperature in celsius")
mfv.fuzzify(22.5)

plt.legend()  # optionally show the legend i.e. "Temperature"
plt.show()

And the output will be:

Others shapes of membership functions can be created such as:

Code for this visualization is available at /lfa_toolbox/examples/readme_examples.py.

Linguistic variable

TODO hand-crafted lv + three points lv + p points lv + screenshot viewer

You can create and visualize linguistic variables by specifing labels and membership functions for each label.

from matplotlib import pyplot as plt
from lfa_toolbox.core.lv.linguistic_variable import LinguisticVariable
from lfa_toolbox.view.lv_viewer import LinguisticVariableViewer

fig, axs = plt.subplots(3, figsize=(12, 8))

for ax in axs:
lv_temp = LinguisticVariable(
    name="temperature",
    ling_values_dict={
        "cold": LinPWMF([17, 1], [20, 0]),
        "warm": LinPWMF([17, 0], [20, 1], [26, 1], [29, 0]),
        "hot": LinPWMF([26, 0], [29, 1]),
    },
)
viewer = LinguisticVariableViewer(lv_temp, ax=ax)
viewer.fuzzify(26.6)
viewer.fuzzify(21.8)

fig.tight_layout()
plt.show()

You can also use the class PPointsLV to create interpretable more conveniently.

from matplotlib import pyplot as plt
from lfa_toolbox.view.lv_viewer import LinguisticVariableViewer
from lfa_toolbox.core.lv.p_points_lv import PPointsLV

fig, ax = plt.subplots()
# PPointsLV helps you create a linguistic variable that is human
# interpretable and generate automatically fuzzy labels for you.
lv = PPointsLV("Github stars", [0, 50, 300, 1000])
LinguisticVariableViewer(lv, ax=ax)

plt.show()

Fuzzy System

You can create an entire fuzzy system by hand. LFA Toolbox supports Singleton and Mamdani fuzzy systems, default rule and NOT conditions. You can see how it works by looking at the examples here /lfa_toolbox/examples.

Take a look at the car_problem_slides folder. It shows a problem where we want to regulate the push of a pedal to maintain the speed of a car. It starts with a Mamdani fuzzy system and gradually simplifies it by first replacing it with a Singleton fuzzy system, then by adding a default fuzzy rule and finally by using NOT conditions.

Integration with Trefle

Trefle is a scikit-learn compatible estimator implementing the FuzzyCoCo algorithm that uses a cooperative coevolution algorithm to find and build interpretable fuzzy systems.

See Trefle's Github to learn how to install it.

Basically after running Trefle you can save the best fuzzy system and fine tune it with LFA Toolbox. For example you can change the values of the membership functions or remove a too specific fuzzy rule.

See /lfa_toolbox/examples/import_a_trefle_fuzzy_system.py example.

Deployment and Tests

Both documentations are available in the docs folder.

Credits

• Gary Marigliano (developper)
• Carlos Andrés PEÑA REYES (project manager)
• CI4CB Team