File size: 6,106 Bytes
a402b9b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 | # Adapted from https://github.com/deepseek-ai/DeepSeek-Math/blob/main/evaluation/eval/eval_utils.py
from math import isclose
import regex
from sympy import N, simplify
from sympy.parsing.latex import parse_latex
from sympy.parsing.sympy_parser import parse_expr
def parse_digits(num):
# format: 234.23 || 23%
num = regex.sub(",", "", str(num))
try:
return float(num)
except:
if num.endswith("%"):
num = num[:-1]
if num.endswith("\\"):
num = num[:-1]
try:
return float(num) / 100
except:
pass
return None
def is_digit(num):
# paired with parse_digits
return parse_digits(num) is not None
def symbolic_equal(a, b):
def _parse(s):
for f in [parse_latex, parse_expr]:
try:
return f(s)
except:
pass
return s
a = _parse(a)
b = _parse(b)
try:
if simplify(a - b) == 0:
return True
except:
pass
try:
if isclose(N(a), N(b), abs_tol=1e-3):
return True
except:
pass
return False
def math_equal(prediction, reference, include_percentage=True, is_close=True):
"""
Exact match of math if and only if:
1. numerical equal: both can convert to float and are equal
2. symbolic equal: both can convert to sympy expression and are equal
"""
if str(prediction) == str(reference):
return True
try: # 1. numerical equal
if is_digit(prediction) and is_digit(reference):
prediction = parse_digits(prediction)
reference = parse_digits(reference)
# number questions
if include_percentage:
gt_result = [reference / 100, reference, reference * 100]
else:
gt_result = [reference]
for item in gt_result:
try:
if is_close:
if isclose(item, prediction, abs_tol=1e-3):
return True
else:
if item == prediction:
return True
except Exception:
continue
return False
except:
pass
if not prediction and prediction not in [0, False]:
return False
# 2. symbolic equal
reference = str(reference).strip()
prediction = str(prediction).strip()
if (
regex.match(r"(\(|\[).+(\)|\])", prediction) is not None
and regex.match(r"(\(|\[).+(\)|\])", reference) is not None
):
pred_parts = prediction[1:-1].split(",")
ref_parts = reference[1:-1].split(",")
if len(pred_parts) == len(ref_parts):
if all(
[
math_equal(
pred_parts[i], ref_parts[i], include_percentage, is_close
)
for i in range(len(pred_parts))
]
):
return True
# Add back matrix comparison
if (
(
prediction.startswith("\\begin{pmatrix}")
or prediction.startswith("\\begin{bmatrix}")
)
and (
prediction.endswith("\\end{pmatrix}")
or prediction.endswith("\\end{bmatrix}")
)
and (
reference.startswith("\\begin{pmatrix}")
or reference.startswith("\\begin{bmatrix}")
)
and (
reference.endswith("\\end{pmatrix}") or reference.endswith("\\end{bmatrix}")
)
):
pred_lines = [
line.strip()
for line in prediction[
len("\\begin{pmatrix}") : -len("\\end{pmatrix}")
].split("\\\\")
if line.strip()
]
ref_lines = [
line.strip()
for line in reference[
len("\\begin{pmatrix}") : -len("\\end{pmatrix}")
].split("\\\\")
if line.strip()
]
matched = True
if len(pred_lines) == len(ref_lines):
for pred_line, ref_line in zip(pred_lines, ref_lines):
pred_parts = pred_line.split("&")
ref_parts = ref_line.split("&")
if len(pred_parts) == len(ref_parts):
if not all(
[
math_equal(
pred_parts[i],
ref_parts[i],
include_percentage,
is_close,
)
for i in range(len(pred_parts))
]
):
matched = False
break
else:
matched = False
if not matched:
break
else:
matched = False
if matched:
return True
# Add back equation comparison
if prediction.count("=") == 1 and reference.count("=") == 1:
pred = prediction.split("=")
pred = f"{pred[0].strip()} - ({pred[1].strip()})"
ref = reference.split("=")
ref = f"{ref[0].strip()} - ({ref[1].strip()})"
if symbolic_equal(pred, ref) or symbolic_equal(f"-({pred})", ref):
return True
elif (
prediction.count("=") == 1
and len(prediction.split("=")[0].strip()) <= 2
and "=" not in reference
):
if math_equal(
prediction.split("=")[1], reference, include_percentage, is_close
):
return True
elif (
reference.count("=") == 1
and len(reference.split("=")[0].strip()) <= 2
and "=" not in prediction
):
if math_equal(
prediction, reference.split("=")[1], include_percentage, is_close
):
return True
# symbolic equal with sympy
if symbolic_equal(prediction, reference):
return True
return False
|