使用正则表达式和极坐标高效解析公式

我正在尝试解析一系列数学公式，需要使用 Python 中的 Polars 高效地提取变量名称。Polars 中的正则表达式支持似乎有限，尤其是对于环视断言。有没有一种简单、有效的方法来解析公式中的符号？

这是我的代码片段：

import re
import polars as pl

# Define the regex pattern
FORMULA_DECODER = r"\b[A-Za-z][A-Za-z_0-9_]*\b(?!\()"
# \b          # Assert a word boundary to ensure matching at the beginning of a word
# [A-Za-z]    # Match an uppercase or lowercase letter at the start
# [A-Za-z0-9_]* # Match following zero or more occurrences of valid characters (letters, digits, or underscores)
# \b          # Assert a word boundary to ensure matching at the end of a word
# (?!\()      # Negative lookahead to ensure the match is not followed by an open parenthesis (indicating a function)

# Sample formulas
formulas = ["3*sin(x1+x2)+A_0",
            "ab*exp(2*x)"]

# expected result
pl.Series(formulas).map_elements(lambda formula: re.findall(FORMULA_DECODER, formula), return_dtype=pl.List(pl.String))
# Series: '' [list[str]]
# [
#   ["x1", "x2", "A_0"]
#   ["ab", "x"]
# ]

# Polars does not support this regex pattern
pl.Series(formulas).str.extract_all(FORMULA_DECODER)
# ComputeError: regex error: regex parse error:
#     \b[A-Za-z][A-Za-z_0-9_]*\b(?!\()
#                               ^^^
# error: look-around, including look-ahead and look-behind, is not supported

编辑这里有一个小的基准：

import random
import string
import re
import polars as pl

def generate_symbol():
    """Generate random symbol of length 1-3."""
    characters = string.ascii_lowercase + string.ascii_uppercase
    return ''.join(random.sample(characters, random.randint(1, 3)))

def generate_formula():
    """Generate random formula with 2-5 unique symbols."""
    op = ['+', '-', '*', '/']
    return ''.join([generate_symbol()+random.choice(op) for _ in range(random.randint(2, 6))])[:-1]


def generate_formulas(num_formulas):
    """Generate random formulas."""
    return [generate_formula() for _ in range(num_formulas)]

# Sample formulas
# formulas = ["3*sin(x1+x2)+(A_0+B)",
#             "ab*exp(2*x)"]

def parse_baseline(formulas):
    """Baseline serves as performance reference. It will not detect function names."""
    FORMULA_DECODER_NO_LOOKAHEAD = r"\b[A-Za-z][A-Za-z_0-9_]*\b\(?"
    return pl.Series(formulas).str.extract_all(FORMULA_DECODER_NO_LOOKAHEAD)

def parse_lookahead(formulas):
    FORMULA_DECODER = r"\b[A-Za-z][A-Za-z_0-9_]*\b(?!\()"
    return pl.Series(formulas).map_elements(lambda formula: re.findall(FORMULA_DECODER, formula), return_dtype=pl.List(pl.String))

def parse_no_lookahead_and_filter(formulas):
    FORMULA_DECODER_NO_LOOKAHEAD = r"\b[A-Za-z][A-Za-z_0-9_]*\b\(?"
    return (
        pl.Series(formulas)
        .str.extract_all(FORMULA_DECODER_NO_LOOKAHEAD)
        # filter for matches not containing an open parenthesis
        .list.eval(pl.element().filter(~pl.element().str.contains("(", literal=True)))
    )

formulas = generate_formulas(1000)
%timeit parse_lookahead(formulas)
%timeit parse_no_lookahead_and_filter(formulas)
%timeit parse_baseline(formulas)
# 10.7 ms ± 387 μs per loop (mean ± std. dev. of 7 runs, 100 loops each)
# 1.31 ms ± 76.1 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)
# 708 μs ± 6.43 μs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)