test_hf_argparser.py 14.3 KB
Newer Older
Sylvain Gugger's avatar
Sylvain Gugger committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
# Copyright 2020 The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

15
import argparse
16
17
import json
import os
18
import sys
19
import tempfile
20
21
22
23
import unittest
from argparse import Namespace
from dataclasses import dataclass, field
from enum import Enum
24
from pathlib import Path
25
from typing import List, Optional
26

27
import yaml
28

Julien Chaumond's avatar
Julien Chaumond committed
29
from transformers import HfArgumentParser, TrainingArguments
30
31
32
33
34
35
36
37
38
from transformers.hf_argparser import make_choice_type_function, string_to_bool


try:
    # For Python versions <3.8, Literal is not in typing: https://peps.python.org/pep-0586/
    from typing import Literal
except ImportError:
    # For Python 3.7
    from typing_extensions import Literal
39

40
41
42
43
# Since Python 3.10, we can use the builtin `|` operator for Union types
# See PEP 604: https://peps.python.org/pep-0604
is_python_no_less_than_3_10 = sys.version_info >= (3, 10)

44

45
46
47
48
def list_field(default=None, metadata=None):
    return field(default_factory=lambda: default, metadata=metadata)


49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
@dataclass
class BasicExample:
    foo: int
    bar: float
    baz: str
    flag: bool


@dataclass
class WithDefaultExample:
    foo: int = 42
    baz: str = field(default="toto", metadata={"help": "help message"})


@dataclass
class WithDefaultBoolExample:
    foo: bool = False
    baz: bool = True
67
    opt: Optional[bool] = None
68
69
70
71
72
73
74


class BasicEnum(Enum):
    titi = "titi"
    toto = "toto"


75
76
77
78
79
80
class MixedTypeEnum(Enum):
    titi = "titi"
    toto = "toto"
    fourtytwo = 42


81
82
@dataclass
class EnumExample:
83
84
85
86
    foo: BasicEnum = "toto"

    def __post_init__(self):
        self.foo = BasicEnum(self.foo)
87
88


89
90
91
92
93
94
95
96
@dataclass
class MixedTypeEnumExample:
    foo: MixedTypeEnum = "toto"

    def __post_init__(self):
        self.foo = MixedTypeEnum(self.foo)


97
98
99
100
101
@dataclass
class OptionalExample:
    foo: Optional[int] = None
    bar: Optional[float] = field(default=None, metadata={"help": "help message"})
    baz: Optional[str] = None
102
103
104
105
106
107
108
109
110
111
    ces: Optional[List[str]] = list_field(default=[])
    des: Optional[List[int]] = list_field(default=[])


@dataclass
class ListExample:
    foo_int: List[int] = list_field(default=[])
    bar_int: List[int] = list_field(default=[1, 2, 3])
    foo_str: List[str] = list_field(default=["Hallo", "Bonjour", "Hello"])
    foo_float: List[float] = list_field(default=[0.1, 0.2, 0.3])
112
113


114
115
116
117
118
119
120
121
122
123
@dataclass
class RequiredExample:
    required_list: List[int] = field()
    required_str: str = field()
    required_enum: BasicEnum = field()

    def __post_init__(self):
        self.required_enum = BasicEnum(self.required_enum)


124
125
126
127
128
129
130
131
132
@dataclass
class StringLiteralAnnotationExample:
    foo: int
    required_enum: "BasicEnum" = field()
    opt: "Optional[bool]" = None
    baz: "str" = field(default="toto", metadata={"help": "help message"})
    foo_str: "List[str]" = list_field(default=["Hallo", "Bonjour", "Hello"])


133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
if is_python_no_less_than_3_10:

    @dataclass
    class WithDefaultBoolExamplePep604:
        foo: bool = False
        baz: bool = True
        opt: bool | None = None

    @dataclass
    class OptionalExamplePep604:
        foo: int | None = None
        bar: float | None = field(default=None, metadata={"help": "help message"})
        baz: str | None = None
        ces: list[str] | None = list_field(default=[])
        des: list[int] | None = list_field(default=[])


150
class HfArgumentParserTest(unittest.TestCase):
151
    def argparsersEqual(self, a: argparse.ArgumentParser, b: argparse.ArgumentParser):
152
153
154
155
156
157
158
        """
        Small helper to check pseudo-equality of parsed arguments on `ArgumentParser` instances.
        """
        self.assertEqual(len(a._actions), len(b._actions))
        for x, y in zip(a._actions, b._actions):
            xx = {k: v for k, v in vars(x).items() if k != "container"}
            yy = {k: v for k, v in vars(y).items() if k != "container"}
159
160
161
162
163
164
165
166

            # Choices with mixed type have custom function as "type"
            # So we need to compare results directly for equality
            if xx.get("choices", None) and yy.get("choices", None):
                for expected_choice in yy["choices"] + xx["choices"]:
                    self.assertEqual(xx["type"](expected_choice), yy["type"](expected_choice))
                del xx["type"], yy["type"]

167
168
169
170
171
172
173
174
175
            self.assertEqual(xx, yy)

    def test_basic(self):
        parser = HfArgumentParser(BasicExample)

        expected = argparse.ArgumentParser()
        expected.add_argument("--foo", type=int, required=True)
        expected.add_argument("--bar", type=float, required=True)
        expected.add_argument("--baz", type=str, required=True)
176
        expected.add_argument("--flag", type=string_to_bool, default=False, const=True, nargs="?")
177
178
        self.argparsersEqual(parser, expected)

179
180
181
182
        args = ["--foo", "1", "--baz", "quux", "--bar", "0.5"]
        (example,) = parser.parse_args_into_dataclasses(args, look_for_args_file=False)
        self.assertFalse(example.flag)

183
184
185
186
187
188
189
190
191
192
    def test_with_default(self):
        parser = HfArgumentParser(WithDefaultExample)

        expected = argparse.ArgumentParser()
        expected.add_argument("--foo", default=42, type=int)
        expected.add_argument("--baz", default="toto", type=str, help="help message")
        self.argparsersEqual(parser, expected)

    def test_with_default_bool(self):
        expected = argparse.ArgumentParser()
193
194
        expected.add_argument("--foo", type=string_to_bool, default=False, const=True, nargs="?")
        expected.add_argument("--baz", type=string_to_bool, default=True, const=True, nargs="?")
195
196
197
        # A boolean no_* argument always has to come after its "default: True" regular counter-part
        # and its default must be set to False
        expected.add_argument("--no_baz", action="store_false", default=False, dest="baz")
198
        expected.add_argument("--opt", type=string_to_bool, default=None)
199

200
201
202
        dataclass_types = [WithDefaultBoolExample]
        if is_python_no_less_than_3_10:
            dataclass_types.append(WithDefaultBoolExamplePep604)
203

204
205
206
        for dataclass_type in dataclass_types:
            parser = HfArgumentParser(dataclass_type)
            self.argparsersEqual(parser, expected)
207

208
209
            args = parser.parse_args([])
            self.assertEqual(args, Namespace(foo=False, baz=True, opt=None))
210

211
212
            args = parser.parse_args(["--foo", "--no_baz"])
            self.assertEqual(args, Namespace(foo=True, baz=False, opt=None))
213

214
215
216
217
218
219
220
221
            args = parser.parse_args(["--foo", "--baz"])
            self.assertEqual(args, Namespace(foo=True, baz=True, opt=None))

            args = parser.parse_args(["--foo", "True", "--baz", "True", "--opt", "True"])
            self.assertEqual(args, Namespace(foo=True, baz=True, opt=True))

            args = parser.parse_args(["--foo", "False", "--baz", "False", "--opt", "False"])
            self.assertEqual(args, Namespace(foo=False, baz=False, opt=False))
222
223

    def test_with_enum(self):
224
        parser = HfArgumentParser(MixedTypeEnumExample)
225
226

        expected = argparse.ArgumentParser()
227
228
229
230
231
232
        expected.add_argument(
            "--foo",
            default="toto",
            choices=["titi", "toto", 42],
            type=make_choice_type_function(["titi", "toto", 42]),
        )
233
234
235
        self.argparsersEqual(parser, expected)

        args = parser.parse_args([])
236
237
        self.assertEqual(args.foo, "toto")
        enum_ex = parser.parse_args_into_dataclasses([])[0]
238
        self.assertEqual(enum_ex.foo, MixedTypeEnum.toto)
239
240

        args = parser.parse_args(["--foo", "titi"])
241
242
        self.assertEqual(args.foo, "titi")
        enum_ex = parser.parse_args_into_dataclasses(["--foo", "titi"])[0]
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
        self.assertEqual(enum_ex.foo, MixedTypeEnum.titi)

        args = parser.parse_args(["--foo", "42"])
        self.assertEqual(args.foo, 42)
        enum_ex = parser.parse_args_into_dataclasses(["--foo", "42"])[0]
        self.assertEqual(enum_ex.foo, MixedTypeEnum.fourtytwo)

    def test_with_literal(self):
        @dataclass
        class LiteralExample:
            foo: Literal["titi", "toto", 42] = "toto"

        parser = HfArgumentParser(LiteralExample)

        expected = argparse.ArgumentParser()
        expected.add_argument(
            "--foo",
            default="toto",
            choices=("titi", "toto", 42),
            type=make_choice_type_function(["titi", "toto", 42]),
        )
        self.argparsersEqual(parser, expected)

        args = parser.parse_args([])
        self.assertEqual(args.foo, "toto")

        args = parser.parse_args(["--foo", "titi"])
        self.assertEqual(args.foo, "titi")

        args = parser.parse_args(["--foo", "42"])
        self.assertEqual(args.foo, 42)
274

275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
    def test_with_list(self):
        parser = HfArgumentParser(ListExample)

        expected = argparse.ArgumentParser()
        expected.add_argument("--foo_int", nargs="+", default=[], type=int)
        expected.add_argument("--bar_int", nargs="+", default=[1, 2, 3], type=int)
        expected.add_argument("--foo_str", nargs="+", default=["Hallo", "Bonjour", "Hello"], type=str)
        expected.add_argument("--foo_float", nargs="+", default=[0.1, 0.2, 0.3], type=float)

        self.argparsersEqual(parser, expected)

        args = parser.parse_args([])
        self.assertEqual(
            args,
            Namespace(foo_int=[], bar_int=[1, 2, 3], foo_str=["Hallo", "Bonjour", "Hello"], foo_float=[0.1, 0.2, 0.3]),
        )

        args = parser.parse_args("--foo_int 1 --bar_int 2 3 --foo_str a b c --foo_float 0.1 0.7".split())
        self.assertEqual(args, Namespace(foo_int=[1], bar_int=[2, 3], foo_str=["a", "b", "c"], foo_float=[0.1, 0.7]))

295
296
297
298
299
    def test_with_optional(self):
        expected = argparse.ArgumentParser()
        expected.add_argument("--foo", default=None, type=int)
        expected.add_argument("--bar", default=None, type=float, help="help message")
        expected.add_argument("--baz", default=None, type=str)
300
301
        expected.add_argument("--ces", nargs="+", default=[], type=str)
        expected.add_argument("--des", nargs="+", default=[], type=int)
302

303
304
305
306
307
308
309
310
311
312
313
        dataclass_types = [OptionalExample]
        if is_python_no_less_than_3_10:
            dataclass_types.append(OptionalExamplePep604)

        for dataclass_type in dataclass_types:
            parser = HfArgumentParser(dataclass_type)

            self.argparsersEqual(parser, expected)

            args = parser.parse_args([])
            self.assertEqual(args, Namespace(foo=None, bar=None, baz=None, ces=[], des=[]))
314

315
316
            args = parser.parse_args("--foo 12 --bar 3.14 --baz 42 --ces a b c --des 1 2 3".split())
            self.assertEqual(args, Namespace(foo=12, bar=3.14, baz="42", ces=["a", "b", "c"], des=[1, 2, 3]))
317

318
319
320
321
322
323
    def test_with_required(self):
        parser = HfArgumentParser(RequiredExample)

        expected = argparse.ArgumentParser()
        expected.add_argument("--required_list", nargs="+", type=int, required=True)
        expected.add_argument("--required_str", type=str, required=True)
324
325
326
327
328
329
        expected.add_argument(
            "--required_enum",
            type=make_choice_type_function(["titi", "toto"]),
            choices=["titi", "toto"],
            required=True,
        )
330
331
        self.argparsersEqual(parser, expected)

332
333
334
335
336
    def test_with_string_literal_annotation(self):
        parser = HfArgumentParser(StringLiteralAnnotationExample)

        expected = argparse.ArgumentParser()
        expected.add_argument("--foo", type=int, required=True)
337
338
339
340
341
342
        expected.add_argument(
            "--required_enum",
            type=make_choice_type_function(["titi", "toto"]),
            choices=["titi", "toto"],
            required=True,
        )
343
344
345
346
347
        expected.add_argument("--opt", type=string_to_bool, default=None)
        expected.add_argument("--baz", default="toto", type=str, help="help message")
        expected.add_argument("--foo_str", nargs="+", default=["Hallo", "Bonjour", "Hello"], type=str)
        self.argparsersEqual(parser, expected)

348
349
350
351
352
353
354
355
356
357
358
359
360
361
    def test_parse_dict(self):
        parser = HfArgumentParser(BasicExample)

        args_dict = {
            "foo": 12,
            "bar": 3.14,
            "baz": "42",
            "flag": True,
        }

        parsed_args = parser.parse_dict(args_dict)[0]
        args = BasicExample(**args_dict)
        self.assertEqual(parsed_args, args)

362
363
364
365
366
367
368
369
370
371
372
373
374
    def test_parse_dict_extra_key(self):
        parser = HfArgumentParser(BasicExample)

        args_dict = {
            "foo": 12,
            "bar": 3.14,
            "baz": "42",
            "flag": True,
            "extra": 42,
        }

        self.assertRaises(ValueError, parser.parse_dict, args_dict, allow_extra_keys=False)

375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
    def test_parse_json(self):
        parser = HfArgumentParser(BasicExample)

        args_dict_for_json = {
            "foo": 12,
            "bar": 3.14,
            "baz": "42",
            "flag": True,
        }
        with tempfile.TemporaryDirectory() as tmp_dir:
            temp_local_path = os.path.join(tmp_dir, "temp_json")
            os.mkdir(temp_local_path)
            with open(temp_local_path + ".json", "w+") as f:
                json.dump(args_dict_for_json, f)
            parsed_args = parser.parse_yaml_file(Path(temp_local_path + ".json"))[0]

        args = BasicExample(**args_dict_for_json)
        self.assertEqual(parsed_args, args)

    def test_parse_yaml(self):
        parser = HfArgumentParser(BasicExample)

        args_dict_for_yaml = {
            "foo": 12,
            "bar": 3.14,
            "baz": "42",
            "flag": True,
        }
        with tempfile.TemporaryDirectory() as tmp_dir:
            temp_local_path = os.path.join(tmp_dir, "temp_yaml")
            os.mkdir(temp_local_path)
            with open(temp_local_path + ".yaml", "w+") as f:
                yaml.dump(args_dict_for_yaml, f)
            parsed_args = parser.parse_yaml_file(Path(temp_local_path + ".yaml"))[0]
        args = BasicExample(**args_dict_for_yaml)
        self.assertEqual(parsed_args, args)

412
413
414
    def test_integration_training_args(self):
        parser = HfArgumentParser(TrainingArguments)
        self.assertIsNotNone(parser)