Find a combination from inputs

[Julien-Livet]

I discovered PySR and I found that it is a powerful tool, thanks for this work!

I have an example that produces me the solution 0 but 0 is not in the inputs, the solution should be the type of x1+x2 (with a multiplicative coefficient). It is for that I wrote my own symbolic regression tool in Python, it is available here: https://github.com/Julien-Livet/SymbolicRegression. You are welcome for some feedback ;-) .

import math
import numpy as np
import pysr
import random

n = 100
xmin = -5
xmax = 10
x1 = (xmax - xmin) * np.random.rand(n) + xmin
x2 = -x1

model = pysr.PySRRegressor(
    model_selection="best",
    niterations=100,
    elementwise_loss="loss(x, y) = (x - y)^2",
    maxsize=20,
    verbosity=1,
)

model.fit(np.transpose([x1, x2]), np.zeros(len(x1)), variable_names = ["x1", "x2"])

import operator
import sr
import sympy

model = sr.SR(niterations = 1,
              binary_operators = {"+": (operator.add, operator.add)},
              avoided_expr = [sympy.sympify("0")],
              foundBreak = True)

model.fit([x1, x2], np.zeros(len(x1)), ["x1", "x2"])

assert(len(model.bestExpressions) == 1)
print(model.bestExpressions[0][0])
assert(sr.sym_expr_eq(sympy.expand(model.bestExpressions[0][0]), sympy.sympify("a*(x1+x2)")))