Optimize float const in sympy expression

opt.py

import numpy as np
import pandas as pd
import sympy as sp
import scipy.optimize as opt


def get_constants(expr):
    const_symbols = []

    def get_constants_helper(expr):
        if isinstance(expr, sp.Float):
            const_symbols.append(float(expr))
        if isinstance(expr, (sp.Symbol, sp.Number)):
            return
        if isinstance(expr, sp.Expr):
            for arg in expr.args:
                get_constants_helper(arg)

    get_constants_helper(expr)
    return const_symbols


def optimize_constants(expr, X, y):
    const_count = 0
    const_symbols = []

    def replace_constants_with_symbols(expr):
        nonlocal const_count, const_symbols

        if isinstance(expr, sp.core.Float):
            const_symbols.append(sp.Symbol('c' + str(const_count)))
            const_count += 1
            return const_symbols[-1]
        if isinstance(expr, (sp.Symbol, sp.Number)):
            return expr
        if isinstance(expr, sp.Expr):
            args = []
            for arg in expr.args:
                args.append(replace_constants_with_symbols(arg))
            return type(expr)(*args)
        return expr

    def objective(coeffs):
        y_pred = np.array(
            f(*[X.iloc[:, i] for i in range(len(variables))], *coeffs)).reshape(-1, 1)
        r = np.mean((y - y_pred)**2, axis=0)
        return r

    expr_with_symbols = replace_constants_with_symbols(expr)

    f = sp.lambdify(list(expr.free_symbols) + const_symbols,
                    expr_with_symbols, "numpy")

    variables = list(expr.free_symbols)
    init_vals = get_constants(expr)
    result = opt.minimize(objective, init_vals)

    # Substitute optimal constants back into expression
    for i, c in enumerate(const_symbols):
        expr_with_symbols = expr_with_symbols.subs(c, result.x[i])

    return expr_with_symbols