File size: 11,240 Bytes
05c9ac2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import pytest
import yaml


from mlagents.trainers.exception import TrainerConfigError, TrainerConfigWarning
from mlagents.trainers.environment_parameter_manager import EnvironmentParameterManager
from mlagents.trainers.settings import (
    RunOptions,
    UniformSettings,
    GaussianSettings,
    ConstantSettings,
    CompletionCriteriaSettings,
)


test_sampler_config_yaml = """
environment_parameters:
  param_1:
    sampler_type: uniform
    sampler_parameters:
      min_value: 0.5
      max_value: 10
"""


def test_sampler_conversion():
    run_options = RunOptions.from_dict(yaml.safe_load(test_sampler_config_yaml))
    assert run_options.environment_parameters is not None
    assert "param_1" in run_options.environment_parameters
    lessons = run_options.environment_parameters["param_1"].curriculum
    assert len(lessons) == 1
    assert lessons[0].completion_criteria is None
    assert isinstance(lessons[0].value, UniformSettings)
    assert lessons[0].value.min_value == 0.5
    assert lessons[0].value.max_value == 10


test_sampler_and_constant_config_yaml = """
environment_parameters:
  param_1:
    sampler_type: gaussian
    sampler_parameters:
      mean: 4
      st_dev: 5
  param_2: 20
"""


def test_sampler_and_constant_conversion():
    run_options = RunOptions.from_dict(
        yaml.safe_load(test_sampler_and_constant_config_yaml)
    )
    assert "param_1" in run_options.environment_parameters
    assert "param_2" in run_options.environment_parameters
    lessons_1 = run_options.environment_parameters["param_1"].curriculum
    lessons_2 = run_options.environment_parameters["param_2"].curriculum
    # gaussian
    assert isinstance(lessons_1[0].value, GaussianSettings)
    assert lessons_1[0].value.mean == 4
    assert lessons_1[0].value.st_dev == 5
    # constant
    assert isinstance(lessons_2[0].value, ConstantSettings)
    assert lessons_2[0].value.value == 20


test_curriculum_config_yaml = """
environment_parameters:
    param_1:
      curriculum:
          - name: Lesson1
            completion_criteria:
                measure: reward
                behavior: fake_behavior
                threshold: 30
                min_lesson_length: 100
                require_reset: true
            value: 1
          - name: Lesson2
            completion_criteria:
                measure: reward
                behavior: fake_behavior
                threshold: 60
                min_lesson_length: 100
                require_reset: false
            value: 2
          - name: Lesson3
            value:
                sampler_type: uniform
                sampler_parameters:
                    min_value: 1
                    max_value: 3
"""


def test_curriculum_conversion():
    run_options = RunOptions.from_dict(yaml.safe_load(test_curriculum_config_yaml))
    assert "param_1" in run_options.environment_parameters
    lessons = run_options.environment_parameters["param_1"].curriculum
    assert len(lessons) == 3
    # First lesson
    lesson = lessons[0]
    assert lesson.completion_criteria is not None
    assert (
        lesson.completion_criteria.measure
        == CompletionCriteriaSettings.MeasureType.REWARD
    )
    assert lesson.completion_criteria.behavior == "fake_behavior"
    assert lesson.completion_criteria.threshold == 30.0
    assert lesson.completion_criteria.min_lesson_length == 100
    assert lesson.completion_criteria.require_reset
    assert isinstance(lesson.value, ConstantSettings)
    assert lesson.value.value == 1
    # Second lesson
    lesson = lessons[1]
    assert lesson.completion_criteria is not None
    assert (
        lesson.completion_criteria.measure
        == CompletionCriteriaSettings.MeasureType.REWARD
    )
    assert lesson.completion_criteria.behavior == "fake_behavior"
    assert lesson.completion_criteria.threshold == 60.0
    assert lesson.completion_criteria.min_lesson_length == 100
    assert not lesson.completion_criteria.require_reset
    assert isinstance(lesson.value, ConstantSettings)
    assert lesson.value.value == 2
    # Last lesson
    lesson = lessons[2]
    assert lesson.completion_criteria is None
    assert isinstance(lesson.value, UniformSettings)
    assert lesson.value.min_value == 1
    assert lesson.value.max_value == 3


test_bad_curriculum_no_competion_criteria_config_yaml = """
environment_parameters:
    param_1:
      curriculum:
          - name: Lesson1
            completion_criteria:
                measure: reward
                behavior: fake_behavior
                threshold: 30
                min_lesson_length: 100
                require_reset: true
            value: 1
          - name: Lesson2
            value: 2
          - name: Lesson3
            value:
                sampler_type: uniform
                sampler_parameters:
                    min_value: 1
                    max_value: 3
"""


test_bad_curriculum_all_competion_criteria_config_yaml = """
environment_parameters:
    param_1:
      curriculum:
          - name: Lesson1
            completion_criteria:
                measure: reward
                behavior: fake_behavior
                threshold: 30
                min_lesson_length: 100
                require_reset: true
            value: 1
          - name: Lesson2
            completion_criteria:
                measure: reward
                behavior: fake_behavior
                threshold: 30
                min_lesson_length: 100
                require_reset: true
            value: 2
          - name: Lesson3
            completion_criteria:
                measure: reward
                behavior: fake_behavior
                threshold: 30
                min_lesson_length: 100
                require_reset: true
            value:
                sampler_type: uniform
                sampler_parameters:
                    min_value: 1
                    max_value: 3
"""


def test_curriculum_raises_no_completion_criteria_conversion():
    with pytest.raises(TrainerConfigError):
        RunOptions.from_dict(
            yaml.safe_load(test_bad_curriculum_no_competion_criteria_config_yaml)
        )


def test_curriculum_raises_all_completion_criteria_conversion():
    with pytest.warns(TrainerConfigWarning):
        run_options = RunOptions.from_dict(
            yaml.safe_load(test_bad_curriculum_all_competion_criteria_config_yaml)
        )

    param_manager = EnvironmentParameterManager(
        run_options.environment_parameters, 1337, False
    )
    assert param_manager.update_lessons(
        trainer_steps={"fake_behavior": 500},
        trainer_max_steps={"fake_behavior": 1000},
        trainer_reward_buffer={"fake_behavior": [1000] * 101},
    ) == (True, True)
    assert param_manager.update_lessons(
        trainer_steps={"fake_behavior": 500},
        trainer_max_steps={"fake_behavior": 1000},
        trainer_reward_buffer={"fake_behavior": [1000] * 101},
    ) == (True, True)
    assert param_manager.update_lessons(
        trainer_steps={"fake_behavior": 500},
        trainer_max_steps={"fake_behavior": 1000},
        trainer_reward_buffer={"fake_behavior": [1000] * 101},
    ) == (False, False)
    assert param_manager.get_current_lesson_number() == {"param_1": 2}


test_everything_config_yaml = """
environment_parameters:
    param_1:
      curriculum:
          - name: Lesson1
            completion_criteria:
                measure: reward
                behavior: fake_behavior
                threshold: 30
                min_lesson_length: 100
                require_reset: true
            value: 1
          - name: Lesson2
            completion_criteria:
                measure: progress
                behavior: fake_behavior
                threshold: 0.5
                min_lesson_length: 100
                require_reset: false
            value: 2
          - name: Lesson3
            value:
                sampler_type: uniform
                sampler_parameters:
                    min_value: 1
                    max_value: 3
    param_2:
        sampler_type: gaussian
        sampler_parameters:
            mean: 4
            st_dev: 5
    param_3: 20
"""


def test_create_manager():
    run_options = RunOptions.from_dict(yaml.safe_load(test_everything_config_yaml))
    param_manager = EnvironmentParameterManager(
        run_options.environment_parameters, 1337, False
    )
    assert param_manager.get_minimum_reward_buffer_size("fake_behavior") == 100
    assert param_manager.get_current_lesson_number() == {
        "param_1": 0,
        "param_2": 0,
        "param_3": 0,
    }
    assert param_manager.get_current_samplers() == {
        "param_1": ConstantSettings(seed=1337, value=1),
        "param_2": GaussianSettings(seed=1337 + 3, mean=4, st_dev=5),
        "param_3": ConstantSettings(seed=1337 + 3 + 1, value=20),
    }
    # Not enough episodes completed
    assert param_manager.update_lessons(
        trainer_steps={"fake_behavior": 500},
        trainer_max_steps={"fake_behavior": 1000},
        trainer_reward_buffer={"fake_behavior": [1000] * 99},
    ) == (False, False)
    # Not enough episodes reward
    assert param_manager.update_lessons(
        trainer_steps={"fake_behavior": 500},
        trainer_max_steps={"fake_behavior": 1000},
        trainer_reward_buffer={"fake_behavior": [1] * 101},
    ) == (False, False)
    assert param_manager.update_lessons(
        trainer_steps={"fake_behavior": 500},
        trainer_max_steps={"fake_behavior": 1000},
        trainer_reward_buffer={"fake_behavior": [1000] * 101},
    ) == (True, True)
    assert param_manager.get_current_lesson_number() == {
        "param_1": 1,
        "param_2": 0,
        "param_3": 0,
    }
    param_manager_2 = EnvironmentParameterManager(
        run_options.environment_parameters, 1337, restore=True
    )
    # The use of global status should make it so that the lesson numbers are maintained
    assert param_manager_2.get_current_lesson_number() == {
        "param_1": 1,
        "param_2": 0,
        "param_3": 0,
    }
    # No reset required
    assert param_manager.update_lessons(
        trainer_steps={"fake_behavior": 700},
        trainer_max_steps={"fake_behavior": 1000},
        trainer_reward_buffer={"fake_behavior": [0] * 101},
    ) == (True, False)
    assert param_manager.get_current_samplers() == {
        "param_1": UniformSettings(seed=1337 + 2, min_value=1, max_value=3),
        "param_2": GaussianSettings(seed=1337 + 3, mean=4, st_dev=5),
        "param_3": ConstantSettings(seed=1337 + 3 + 1, value=20),
    }


test_curriculum_no_behavior_yaml = """
environment_parameters:
    param_1:
      curriculum:
          - name: Lesson1
            completion_criteria:
                measure: reward
                threshold: 30
                min_lesson_length: 100
                require_reset: true
            value: 1
          - name: Lesson2
            value: 2
"""


def test_curriculum_no_behavior():
    with pytest.raises(TypeError):
        run_options = RunOptions.from_dict(
            yaml.safe_load(test_curriculum_no_behavior_yaml)
        )
        EnvironmentParameterManager(run_options.environment_parameters, 1337, False)