How to use?

After installing symlearn package, you can use it by importing the essential modules and classes. The following example shows how to use symlearn package in your projects. Example codes and data files can be found in Package Repository.

In this example, we used FFP is used in modeling and forecasting the system of Box-Jenkins gas furnace time series. It starts with importing dataset and splitting it into training and test subsets:

# import dataset and split it into training and test subsets
box_dataset = pd.read_csv('datasets/box-jenkins.csv').to_numpy()
X = box_dataset[:, [0, 1]]
y = box_dataset[:, 2]
t = np.arange(0, 290)
X_train = X[0:201, :]
X_test = X[200:, :]
y_train = y[0:201]
y_test = y[200:]
t_train = t[:201]
t_test = t[200:]

After that symlearn’s core functions and modules must be imported.

# import core dependicies
from symlearn.core.parameters import Parameters
from symlearn.core.errors import sum_of_difference
from symlearn.core.functions import *

After Importing the essential core modules, Parameters must be declared before start of optimizing process.

# set parameters, expressions, and terminals
Parameters.CONSTANTS = [-1, 1]
Parameters.FEATURES = X_train.shape[1]
Parameters.CONSTANTS_TYPE = 'range'
expressions = [Add, Sub, Mul, Div, Pow]
terminals = [Variable, Constant]

A suitable model must be chosen, and its arguments must be satisfied. Here in this example, we used FFP model with the arguments shown in the following code block:

# import error metric, FFP model and initialize it     
from symlearn.models import FFP

model = FFP(pop_size=50,
        max_evaluations=5000,
        initial_min_depth=0,
        initial_max_depth=6,
        min_depth=1,
        max_depth=15,
        error_function=sum_of_difference,
        expressions=expressions,
        terminals=terminals,
        target_error=0,
        verbose=True
        )

Fitting model using fit() method.

model.fit(X_train, y_train)
y_fit = model.predict(X_train)
y_pred = model.predict(X_test)

Evaluations: 5 | Fitness: 14.163321775790926
Evaluations: 68 | Fitness: 8.194608
Evaluations: 823 | Fitness: 5.095621733489572
Evaluations: 1917 | Fitness: 4.3966582987110225
Evaluations: 2850 | Fitness: 3.7636233912595776
Evaluations: 3433 | Fitness: 3.561661510016762
Evaluations: 4148 | Fitness: 3.540966725348731
Evaluations: 5001

Plotting the optimized model

# plotting the optimized model    
import matplotlib.pyplot as plt
ax = plt.axes()
ax.grid(linestyle=':', linewidth=0.5, alpha=1, zorder=1)
line = [None, None, None, None]
line[0], = ax.plot(t_train, y_train, linestyle=':', color='black', linewidth=0.7, zorder=2, label='Targeted')    
line[1], = ax.plot(t_train, y_fit, linestyle='-', color='red', linewidth=0.7, zorder=3, label='Trained')
line[2], = ax.plot(t_test, y_test, linestyle=':', color='black', linewidth=0.7, zorder=2)
line[3], = ax.plot(t_test, y_pred, linestyle='-', color='blue', linewidth=0.7, zorder=3, label='Predicted')
plt.axvline(x=t_test[0], linestyle='-', color='black', linewidth='1')
plt.draw()
plt.legend()
plt.show()

Targeted and Predicted models’ graphs

Exporting the graph of the best model:

model.export_best()

((x2)-(((((x0)+(x1))-(-1.74))-(x1))+((((-1.12)*(1.06))-((x1)+(-3.78)))-(1.41))))

Best models’ tree representation