garrethlee commited on
Commit
a2886f3
·
verified ·
1 Parent(s): 907c17d

Create simple-arithmetic-problems.py

Browse files
Files changed (1) hide show
  1. simple-arithmetic-problems.py +433 -0
simple-arithmetic-problems.py ADDED
@@ -0,0 +1,433 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import random
2
+ from datasets import (
3
+ BuilderConfig,
4
+ SplitGenerator,
5
+ GeneratorBasedBuilder,
6
+ DatasetInfo,
7
+ Value,
8
+ Features,
9
+ )
10
+ from decimal import Decimal
11
+ import yaml
12
+
13
+ SEED = 42
14
+ TEST_SIZE = 0.2
15
+ DIVISION_RESULT_MULTIPLIER = 4
16
+ FLOAT_FLOAT_PROBLEM_PROPORTION = 0.3
17
+
18
+ _CITATION = """\
19
+ @misc{lee2024simplearithmeticproblemsdataset,
20
+ title = {Simple Arithmetic Problems},
21
+ author={Garreth Lee},
22
+ year={2024}
23
+ }
24
+ """
25
+
26
+
27
+ class Operator:
28
+ ADD = "+"
29
+ SUBTRACT = "-"
30
+ MULTIPLY = "*"
31
+ DIVIDE = "/"
32
+
33
+ OPERATORS = [ADD, SUBTRACT, MULTIPLY, DIVIDE]
34
+
35
+ @classmethod
36
+ def is_operator(cls, value):
37
+ return value in cls.OPERATORS
38
+
39
+ @classmethod
40
+ def operator_to_name(cls, value):
41
+ if value == cls.ADD:
42
+ return "add"
43
+ elif value == cls.SUBTRACT:
44
+ return "subtract"
45
+ elif value == cls.MULTIPLY:
46
+ return "multiply"
47
+ elif value == cls.DIVIDE:
48
+ return "divide"
49
+ else:
50
+ raise ValueError(f"Invalid operator: {value}")
51
+
52
+
53
+ class OperationType:
54
+ INT_INT = [False, False]
55
+ INT_FLOAT = [True, False]
56
+ FLOAT_FLOAT = [True, True]
57
+
58
+
59
+ class ArithmeticProblemsConfig:
60
+ def __init__(
61
+ self,
62
+ config_file_path: str,
63
+ ) -> None:
64
+ with open(config_file_path) as f:
65
+ self.config = yaml.safe_load(f)
66
+
67
+
68
+ class ArithmeticProblemsConfig(BuilderConfig):
69
+ def __init__(
70
+ self,
71
+ name: str,
72
+ num_problems: int,
73
+ min_exponent: int,
74
+ max_exponent: int,
75
+ max_rounding_precision: int,
76
+ use_commas: bool = False,
77
+ **kwargs,
78
+ ):
79
+ super().__init__(name=name)
80
+ self.num_problems = num_problems
81
+ self.min_exponent = min_exponent
82
+ self.max_exponent = max_exponent
83
+ self.max_rounding_precision = max_rounding_precision
84
+ self.use_commas = use_commas
85
+ self.kwargs = kwargs
86
+
87
+
88
+ class ArithmeticProblemsDataset(GeneratorBasedBuilder):
89
+ BUILDER_CONFIG_CLASS = ArithmeticProblemsConfig
90
+ FLOAT_ANSWER_ROUNDING_PRECISION = 4
91
+
92
+ BUILDER_CONFIGS = [
93
+ ArithmeticProblemsConfig(
94
+ name="very_easy",
95
+ num_problems=5000,
96
+ min_exponent=0,
97
+ max_exponent=1,
98
+ max_rounding_precision=1,
99
+ ),
100
+ ArithmeticProblemsConfig(
101
+ name="easy",
102
+ num_problems=5000,
103
+ min_exponent=0,
104
+ max_exponent=2,
105
+ max_rounding_precision=2,
106
+ ),
107
+ ArithmeticProblemsConfig(
108
+ name="medium",
109
+ num_problems=5000,
110
+ min_exponent=1,
111
+ max_exponent=3,
112
+ max_rounding_precision=3,
113
+ ),
114
+ ArithmeticProblemsConfig(
115
+ name="hard",
116
+ num_problems=5000,
117
+ min_exponent=1,
118
+ max_exponent=4,
119
+ max_rounding_precision=4,
120
+ ),
121
+ ArithmeticProblemsConfig(
122
+ name="very_hard",
123
+ num_problems=5000,
124
+ min_exponent=2,
125
+ max_exponent=5,
126
+ max_rounding_precision=5,
127
+ ),
128
+ ArithmeticProblemsConfig(
129
+ name="very_easy_use_commas",
130
+ num_problems=5000,
131
+ min_exponent=0,
132
+ max_exponent=1,
133
+ max_rounding_precision=1,
134
+ use_commas=True,
135
+ ),
136
+ ArithmeticProblemsConfig(
137
+ name="easy_use_commas",
138
+ num_problems=5000,
139
+ min_exponent=0,
140
+ max_exponent=2,
141
+ max_rounding_precision=2,
142
+ use_commas=True,
143
+ ),
144
+ ArithmeticProblemsConfig(
145
+ name="medium_use_commas",
146
+ num_problems=5000,
147
+ min_exponent=1,
148
+ max_exponent=3,
149
+ max_rounding_precision=3,
150
+ use_commas=True,
151
+ ),
152
+ ArithmeticProblemsConfig(
153
+ name="hard_use_commas",
154
+ num_problems=5000,
155
+ min_exponent=1,
156
+ max_exponent=4,
157
+ max_rounding_precision=4,
158
+ use_commas=True,
159
+ ),
160
+ ArithmeticProblemsConfig(
161
+ name="very_hard_use_commas",
162
+ num_problems=5000,
163
+ min_exponent=2,
164
+ max_exponent=5,
165
+ max_rounding_precision=5,
166
+ use_commas=True,
167
+ ),
168
+ ]
169
+
170
+ VERSION = "1.0.0"
171
+
172
+ def _info(self):
173
+ return DatasetInfo(
174
+ description="Generate arithmetic problems for use in math tokenization",
175
+ features=Features(
176
+ {
177
+ "question": Value("string"),
178
+ "answer": Value("string"),
179
+ "operator": Value("string"),
180
+ }
181
+ ),
182
+ citation=_CITATION,
183
+ )
184
+
185
+ def _generate_number(
186
+ self, min_val: int, max_val: int, is_float: bool, max_rounding_precision: int
187
+ ) -> float | int:
188
+ """
189
+ Generates a random number within a specified range, either as an integer or float.
190
+
191
+ Args:
192
+ min_val: The minimum value of the range.
193
+ max_val: The maximum value of the range.
194
+ is_float: If true, generates a float
195
+ max_rounding_precision: The maximum precision to use when rounding the number.
196
+
197
+ Returns:
198
+ A random number within the specified range, either as an int or a float.
199
+ """
200
+ if is_float:
201
+ # Round to a random precision between 0 and max_rounding_precision
202
+ return round(
203
+ random.uniform(min_val, max_val),
204
+ random.choice(range(1, max_rounding_precision + 1)),
205
+ )
206
+ else:
207
+ return random.randint(min_val, max_val)
208
+
209
+ def _format_number(self, number: int | float, use_commas: bool = False) -> str:
210
+ """
211
+ Rounds a number to a specified precision, and then formats it as a string.
212
+
213
+ Args:
214
+ number: The number to be formatted.
215
+ use_commas: Whether to include commas as thousand separators.
216
+
217
+ Returns:
218
+ A string representation of the input number, rounded to the specified precision.
219
+ """
220
+ if use_commas:
221
+ return "{:,}".format(number)
222
+ else:
223
+ return str(number)
224
+
225
+ def _construct_equation(
226
+ self,
227
+ operand1: int | float,
228
+ operand2: int | float,
229
+ operator: str,
230
+ use_commas: bool = False,
231
+ ) -> str:
232
+ """Helper function for constructing the string equations."""
233
+
234
+ return "%s %s %s = " % (
235
+ self._format_number(operand1, use_commas),
236
+ operator,
237
+ self._format_number(operand2, use_commas),
238
+ )
239
+
240
+ def create_question_answer_pair(
241
+ self,
242
+ min_value: int,
243
+ max_value: int,
244
+ operator: str,
245
+ use_commas: bool,
246
+ operation_type: list[bool],
247
+ max_rounding_precision: int | None,
248
+ ) -> dict[str, str]:
249
+ """Creates a random question and correct answer pair.
250
+
251
+ Args:
252
+ min_value: The lowest possible random value.
253
+ max_value: The highest possible random value.
254
+ include_decimals: Whether to include float numbers in the generated problems.
255
+ operator: The mathematical operator to use.
256
+ use_commas: Whether to use commas to separate numbers right-to-left.
257
+
258
+ Returns:
259
+ A dictionary containing the equation string and the expected answer.
260
+ """
261
+ if not Operator.is_operator(operator):
262
+ raise ValueError(f"Invalid operator: {operator}")
263
+
264
+ is_float1, is_float2 = operation_type
265
+ operand1 = self._generate_number(
266
+ min_val=min_value,
267
+ max_val=max_value,
268
+ is_float=is_float1,
269
+ max_rounding_precision=max_rounding_precision,
270
+ )
271
+ operand2 = self._generate_number(
272
+ min_val=min_value,
273
+ max_val=max_value,
274
+ is_float=is_float2,
275
+ max_rounding_precision=max_rounding_precision,
276
+ )
277
+
278
+ if operator == Operator.SUBTRACT:
279
+ result = operand1 - operand2
280
+ elif operator == Operator.ADD:
281
+ result = operand1 + operand2
282
+ elif operator == Operator.MULTIPLY:
283
+ result = operand1 * operand2
284
+ else:
285
+ # this prevents a lot of "0" answers from being generated
286
+ if operand1 < operand2 or random.random() < 0.01:
287
+ operand1, operand2 = operand2, operand1
288
+
289
+ if operation_type == OperationType.INT_INT:
290
+ tmp = operand1 / operand2
291
+
292
+ # we scale the temp result up to prevent a lot of "small divisions"
293
+ if tmp < 10:
294
+ tmp *= DIVISION_RESULT_MULTIPLIER
295
+ if max_value > 999:
296
+ tmp *= random.randint(2, 4)
297
+
298
+ # prevents zero division
299
+ operand1 = int(round(tmp)) * operand2
300
+ result = int(operand1 / operand2)
301
+
302
+ elif operation_type == OperationType.INT_FLOAT:
303
+ operand2 = int(operand2)
304
+ tmp = round(
305
+ operand1 / operand2,
306
+ random.randint(1, max_rounding_precision),
307
+ )
308
+
309
+ # we scale the temp result up to prevent a lot of "small divisions"
310
+ if tmp < 10:
311
+ tmp = float(
312
+ Decimal(str(tmp)) * Decimal(str(DIVISION_RESULT_MULTIPLIER))
313
+ )
314
+
315
+ # deals with Python's decimal multiplication precision issue
316
+ operand1 = float(Decimal(str(tmp)) * Decimal(str(operand2)))
317
+ result = tmp
318
+
319
+ else:
320
+ tmp = round(
321
+ operand1 / operand2, random.randint(1, max_rounding_precision)
322
+ )
323
+
324
+ # we scale the temp result up to prevent a lot of "small divisions"
325
+ if tmp < 10:
326
+ tmp = float(
327
+ Decimal(str(tmp)) * Decimal(str(DIVISION_RESULT_MULTIPLIER))
328
+ )
329
+
330
+ # deals with Python's decimal multiplication precision issue
331
+ operand1 = float(Decimal(str(tmp)) * Decimal(str(operand2)))
332
+ result = tmp
333
+
334
+ result = round(result, self.FLOAT_ANSWER_ROUNDING_PRECISION)
335
+
336
+ question = self._construct_equation(
337
+ operand1=operand1,
338
+ operand2=operand2,
339
+ operator=operator,
340
+ use_commas=use_commas,
341
+ )
342
+ answer = self._format_number(result, use_commas)
343
+
344
+ return {"question": question, "answer": answer, "operator": operator}
345
+
346
+ def _split_generators(self, dl_manager, **kwargs) -> list[SplitGenerator]:
347
+ generators = []
348
+
349
+ for operator in Operator.OPERATORS:
350
+ # Create separate splits for each type of number
351
+ for type in ("int", "float"):
352
+ split_name = f"{type}_{Operator.operator_to_name(operator)}"
353
+
354
+ train_generator = SplitGenerator(
355
+ name=split_name + "_train",
356
+ gen_kwargs={
357
+ "num_problems": int(self.config.num_problems * (1 - TEST_SIZE)),
358
+ "min_value": 10**self.config.min_exponent,
359
+ "max_value": 10**self.config.max_exponent,
360
+ "max_rounding_precision": self.config.max_rounding_precision
361
+ if type == "float"
362
+ else None,
363
+ "use_commas": self.config.use_commas,
364
+ "operator": operator,
365
+ },
366
+ )
367
+
368
+ test_generator = SplitGenerator(
369
+ name=split_name + "_test",
370
+ gen_kwargs={
371
+ "num_problems": int(self.config.num_problems * TEST_SIZE),
372
+ "min_value": 10**self.config.min_exponent,
373
+ "max_value": 10**self.config.max_exponent,
374
+ "max_rounding_precision": self.config.max_rounding_precision
375
+ if type == "float"
376
+ else None,
377
+ "use_commas": self.config.use_commas,
378
+ "operator": operator,
379
+ },
380
+ )
381
+
382
+ generators.append(train_generator)
383
+ generators.append(test_generator)
384
+
385
+ return generators
386
+
387
+ def _generate_examples(
388
+ self,
389
+ num_problems,
390
+ min_value,
391
+ max_value,
392
+ max_rounding_precision,
393
+ use_commas,
394
+ operator,
395
+ ):
396
+ def _get_operation_type(current_idx: int):
397
+ # If max_rounding_precision is None, generate only integer problems
398
+ """
399
+ Determines the type of operation (integer-integer, float-float, or integer-float)
400
+ to generate based on the current index and the proportion of float problems.
401
+
402
+ Args:
403
+ current_idx: The current index of the problem being generated.
404
+ num_problems: The total number of problems to generate.
405
+ max_rounding_precision: The maximum rounding precision to use when generating float problems.
406
+
407
+ Returns:
408
+ An OperationType indicating the type of operation to generate.
409
+ """
410
+ if max_rounding_precision is None:
411
+ return OperationType.INT_INT
412
+
413
+ # Otherwise, if the current index is less than the float problem proportion,
414
+ elif current_idx < num_problems * FLOAT_FLOAT_PROBLEM_PROPORTION:
415
+ return OperationType.FLOAT_FLOAT
416
+
417
+ else:
418
+ return OperationType.INT_FLOAT
419
+
420
+ random.seed(SEED)
421
+
422
+ for i in range(num_problems):
423
+ yield (
424
+ str(i),
425
+ self.create_question_answer_pair(
426
+ min_value=min_value,
427
+ max_value=max_value,
428
+ operator=operator,
429
+ use_commas=use_commas,
430
+ operation_type=_get_operation_type(i),
431
+ max_rounding_precision=max_rounding_precision,
432
+ ),
433
+ )