File size: 12,119 Bytes
94e3dda |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 |
## Adapted from https://github.com/maxwbuckley/r2ltokenizing/blob/main/llmarithmetic.py; with modifications
import random
from datasets import (
BuilderConfig,
SplitGenerator,
GeneratorBasedBuilder,
DatasetInfo,
Value,
Features,
)
from decimal import Decimal
import yaml
SEED = 42
TEST_SIZE = 0.2
DIVISION_RESULT_MULTIPLIER = 4
FLOAT_FLOAT_PROBLEM_PROPORTION = 0.3
_CITATION = """\
@misc{lee2024arithmeticproblemsdataset,
title = {Arithmetic Problems},
author={Garreth Lee},
year={2024}
}
"""
class Operator:
ADD = "+"
SUBTRACT = "-"
MULTIPLY = "*"
DIVIDE = "/"
OPERATORS = [ADD, SUBTRACT, MULTIPLY, DIVIDE]
@classmethod
def is_operator(cls, value):
return value in cls.OPERATORS
@classmethod
def operator_to_name(cls, value):
if value == cls.ADD:
return "add"
elif value == cls.SUBTRACT:
return "subtract"
elif value == cls.MULTIPLY:
return "multiply"
elif value == cls.DIVIDE:
return "divide"
else:
raise ValueError(f"Invalid operator: {value}")
class OperationType:
INT_INT = [False, False]
INT_FLOAT = [True, False]
FLOAT_FLOAT = [True, True]
class ArithmeticProblemsConfig(BuilderConfig):
def __init__(
self,
name: str,
num_problems: int,
min_exponent: int,
max_exponent: int,
max_rounding_precision: int,
use_commas: bool = False,
**kwargs,
):
super().__init__(name=name)
self.num_problems = num_problems
self.min_exponent = min_exponent
self.max_exponent = max_exponent
self.max_rounding_precision = max_rounding_precision
self.use_commas = use_commas
self.kwargs = kwargs
class ArithmeticProblemsDataset(GeneratorBasedBuilder):
BUILDER_CONFIG_CLASS = ArithmeticProblemsConfig
FLOAT_ANSWER_ROUNDING_PRECISION = 4
BUILDER_CONFIGS = [
ArithmeticProblemsConfig(
name=f"{i}-digit{'-use-commas' if use_commas else ''}",
num_problems=5000,
min_exponent=i-1,
max_exponent=i,
max_rounding_precision=max(i-1, 10),
use_commas = use_commas,
) for i in range(1, 21) for use_commas in (True, False)
]
VERSION = "1.0.0"
def _info(self):
return DatasetInfo(
description="Generate arithmetic problems for use in math tokenization",
features=Features(
{
"question": Value("string"),
"answer": Value("string"),
"operator": Value("string"),
}
),
citation=_CITATION,
)
def _generate_number(
self, min_val: int, max_val: int, is_float: bool, max_rounding_precision: int
) -> float | int:
"""
Generates a random number within a specified range, either as an integer or float.
Args:
min_val: The minimum value of the range.
max_val: The maximum value of the range.
is_float: If true, generates a float
max_rounding_precision: The maximum precision to use when rounding the number.
Returns:
A random number within the specified range, either as an int or a float.
"""
if is_float:
# Round to a random precision between 0 and max_rounding_precision
return round(
random.uniform(min_val, max_val),
random.choice(range(1, max_rounding_precision + 1)),
)
else:
return random.randint(min_val, max_val)
def _format_number(self, number: int | float, use_commas: bool = False) -> str:
"""
Rounds a number to a specified precision, and then formats it as a string.
Args:
number: The number to be formatted.
use_commas: Whether to include commas as thousand separators.
Returns:
A string representation of the input number, rounded to the specified precision.
"""
if use_commas:
return "{:,}".format(number)
else:
return str(number)
def _construct_equation(
self,
operand1: int | float,
operand2: int | float,
operator: str,
use_commas: bool = False,
) -> str:
"""Helper function for constructing the string equations."""
return "%s %s %s = " % (
self._format_number(operand1, use_commas),
operator,
self._format_number(operand2, use_commas),
)
def create_question_answer_pair(
self,
min_value: int,
max_value: int,
operator: str,
use_commas: bool,
operation_type: list[bool],
max_rounding_precision: int | None,
) -> dict[str, str]:
"""Creates a random question and correct answer pair.
Args:
min_value: The lowest possible random value.
max_value: The highest possible random value.
include_decimals: Whether to include float numbers in the generated problems.
operator: The mathematical operator to use.
use_commas: Whether to use commas to separate numbers right-to-left.
Returns:
A dictionary containing the equation string and the expected answer.
"""
if not Operator.is_operator(operator):
raise ValueError(f"Invalid operator: {operator}")
is_float1, is_float2 = operation_type
operand1 = self._generate_number(
min_val=min_value,
max_val=max_value,
is_float=is_float1,
max_rounding_precision=max_rounding_precision,
)
operand2 = self._generate_number(
min_val=min_value,
max_val=max_value,
is_float=is_float2,
max_rounding_precision=max_rounding_precision,
)
if operator == Operator.SUBTRACT:
result = operand1 - operand2
elif operator == Operator.ADD:
result = operand1 + operand2
elif operator == Operator.MULTIPLY:
result = operand1 * operand2
else:
# this prevents a lot of "0" answers from being generated
if operand1 < operand2 or random.random() < 0.01:
operand1, operand2 = operand2, operand1
if operation_type == OperationType.INT_INT:
tmp = operand1 / operand2
# we scale the temp result up to prevent a lot of "small divisions"
if tmp < 10:
tmp *= DIVISION_RESULT_MULTIPLIER
if max_value > 999:
tmp *= random.randint(2, 4)
# prevents zero division
operand1 = int(round(tmp)) * operand2
result = int(operand1 / operand2)
elif operation_type == OperationType.INT_FLOAT:
operand2 = int(operand2)
tmp = round(
operand1 / operand2,
random.randint(1, max_rounding_precision),
)
# we scale the temp result up to prevent a lot of "small divisions"
if tmp < 10:
tmp = float(
Decimal(str(tmp)) * Decimal(str(DIVISION_RESULT_MULTIPLIER))
)
# deals with Python's decimal multiplication precision issue
operand1 = float(Decimal(str(tmp)) * Decimal(str(operand2)))
result = tmp
else:
tmp = round(
operand1 / operand2, random.randint(1, max_rounding_precision)
)
# we scale the temp result up to prevent a lot of "small divisions"
if tmp < 10:
tmp = float(
Decimal(str(tmp)) * Decimal(str(DIVISION_RESULT_MULTIPLIER))
)
# deals with Python's decimal multiplication precision issue
operand1 = float(Decimal(str(tmp)) * Decimal(str(operand2)))
result = tmp
result = round(result, self.FLOAT_ANSWER_ROUNDING_PRECISION)
question = self._construct_equation(
operand1=operand1,
operand2=operand2,
operator=operator,
use_commas=use_commas,
)
answer = self._format_number(result, use_commas)
return {"question": question, "answer": answer, "operator": operator}
def _split_generators(self, dl_manager, **kwargs) -> list[SplitGenerator]:
generators = []
for operator in Operator.OPERATORS:
# Create separate splits for each type of number
for type in ("int", "float"):
split_name = f"{type}_{Operator.operator_to_name(operator)}"
train_generator = SplitGenerator(
name=split_name + "_train",
gen_kwargs={
"num_problems": int(self.config.num_problems * (1 - TEST_SIZE)),
"min_value": 10**self.config.min_exponent,
"max_value": 10**self.config.max_exponent,
"max_rounding_precision": self.config.max_rounding_precision
if type == "float"
else None,
"use_commas": self.config.use_commas,
"operator": operator,
},
)
test_generator = SplitGenerator(
name=split_name + "_test",
gen_kwargs={
"num_problems": int(self.config.num_problems * TEST_SIZE),
"min_value": 10**self.config.min_exponent,
"max_value": 10**self.config.max_exponent,
"max_rounding_precision": self.config.max_rounding_precision
if type == "float"
else None,
"use_commas": self.config.use_commas,
"operator": operator,
},
)
generators.append(train_generator)
generators.append(test_generator)
return generators
def _generate_examples(
self,
num_problems,
min_value,
max_value,
max_rounding_precision,
use_commas,
operator,
):
def _get_operation_type(current_idx: int):
# If max_rounding_precision is None, generate only integer problems
"""
Determines the type of operation (integer-integer, float-float, or integer-float)
to generate based on the current index and the proportion of float problems.
Args:
current_idx: The current index of the problem being generated.
num_problems: The total number of problems to generate.
max_rounding_precision: The maximum rounding precision to use when generating float problems.
Returns:
An OperationType indicating the type of operation to generate.
"""
if max_rounding_precision is None:
return OperationType.INT_INT
# Otherwise, if the current index is less than the float problem proportion,
elif current_idx < num_problems * FLOAT_FLOAT_PROBLEM_PROPORTION:
return OperationType.FLOAT_FLOAT
else:
return OperationType.INT_FLOAT
random.seed(SEED)
for i in range(num_problems):
yield (
str(i),
self.create_question_answer_pair(
min_value=min_value,
max_value=max_value,
operator=operator,
use_commas=use_commas,
operation_type=_get_operation_type(i),
max_rounding_precision=max_rounding_precision,
),
)
|