forked from fedbiomed/fedbiomed
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtest_training_iterations.py
212 lines (193 loc) · 10.1 KB
/
test_training_iterations.py
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
import unittest
from unittest.mock import MagicMock
import numpy as np
from fedbiomed.common.training_plans._training_iterations import MiniBatchTrainingIterationsAccountant # noqa
reference_num_batches = np.random.randint(low=4, high=15, size=1)[0]
reference_batch_size = np.random.randint(low=1, high=5, size=1)[0]
reference_dataset_size = reference_num_batches*reference_batch_size
mock_training_plan = MagicMock()
mock_training_plan.training_data_loader.__len__.return_value = reference_num_batches
mock_training_plan.training_data_loader.dataset.__len__.return_value = reference_dataset_size
class TestMiniBatchTrainingIterationsAccountant(unittest.TestCase):
def test_min_batch_training_iterations_accountant_01_epochs_num_updates(self):
# Researcher asks for 1 epoch
mock_training_plan.training_args.return_value = {
'epochs': 1,
'batch_maxnum': None,
'num_updates': None,
}
iter_accountant = MiniBatchTrainingIterationsAccountant(mock_training_plan)
self.assertEqual(iter_accountant.epochs, 2)
self.assertEqual(iter_accountant.num_batches_in_last_epoch, 0)
self.assertEqual(iter_accountant.num_batches_per_epoch, reference_num_batches)
# Researcher asks for 1 epoch but batch_maxnum != 0
mock_training_plan.training_args.return_value = {
'epochs': 1,
'batch_maxnum': 3,
'num_updates': None,
}
iter_accountant = MiniBatchTrainingIterationsAccountant(mock_training_plan)
self.assertEqual(iter_accountant.epochs, 2)
self.assertEqual(iter_accountant.num_batches_in_last_epoch, 0)
self.assertEqual(iter_accountant.num_batches_per_epoch, 3)
# Researcher asks for 1 epoch but batch_maxnum == 0
mock_training_plan.training_args.return_value = {
'epochs': 1,
'batch_maxnum': 0,
'num_updates': None,
}
iter_accountant = MiniBatchTrainingIterationsAccountant(mock_training_plan)
self.assertEqual(iter_accountant.epochs, 2)
self.assertEqual(iter_accountant.num_batches_in_last_epoch, 0)
self.assertEqual(iter_accountant.num_batches_per_epoch, reference_num_batches)
# Researcher asks for len_data_loader updates
mock_training_plan.training_args.return_value = {
'epochs': None,
'batch_maxnum': None,
'num_updates': reference_num_batches,
}
iter_accountant = MiniBatchTrainingIterationsAccountant(mock_training_plan)
self.assertEqual(iter_accountant.epochs, 2)
self.assertEqual(iter_accountant.num_batches_in_last_epoch, 0)
self.assertEqual(iter_accountant.num_batches_per_epoch, reference_num_batches)
# Researcher asks for updates < len_data_loader
mock_training_plan.training_args.return_value = {
'epochs': None,
'batch_maxnum': None,
'num_updates': reference_num_batches - 3,
}
iter_accountant = MiniBatchTrainingIterationsAccountant(mock_training_plan)
self.assertEqual(iter_accountant.epochs, 1)
self.assertEqual(iter_accountant.num_batches_in_last_epoch, reference_num_batches - 3)
self.assertEqual(iter_accountant.num_batches_per_epoch, reference_num_batches)
# Researcher asks for updates > len_data_loader
mock_training_plan.training_args.return_value = {
'epochs': None,
'batch_maxnum': None,
'num_updates': 2*reference_num_batches + 3,
}
iter_accountant = MiniBatchTrainingIterationsAccountant(mock_training_plan)
self.assertEqual(iter_accountant.epochs, 3)
self.assertEqual(iter_accountant.num_batches_in_last_epoch, 3)
self.assertEqual(iter_accountant.num_batches_per_epoch, reference_num_batches)
# Precedence rules
mock_training_plan.training_args.return_value = {
'epochs': 42,
'batch_maxnum': 4242,
'num_updates': 2*reference_num_batches + 3,
}
iter_accountant = MiniBatchTrainingIterationsAccountant(mock_training_plan)
self.assertEqual(iter_accountant.epochs, 3)
self.assertEqual(iter_accountant.num_batches_in_last_epoch, 3)
self.assertEqual(iter_accountant.num_batches_per_epoch, reference_num_batches)
def test_mini_batch_training_iterations_accountaint_02_iterations(self):
mock_training_plan.training_args.return_value = {
'epochs': None,
'batch_maxnum': None,
'num_updates': 2*reference_num_batches + 3,
}
iter_accountant = MiniBatchTrainingIterationsAccountant(mock_training_plan)
epoch_counter = 0
tot_steps_counter = 0
for epoch in iter_accountant.iterate_epochs():
epoch_counter += 1
batch_counter = 0
self.assertEqual(epoch_counter, iter_accountant.cur_epoch)
for batch in iter_accountant.iterate_batches():
batch_counter += 1
tot_steps_counter += 1
self.assertEqual(batch_counter, iter_accountant.cur_batch)
self.assertEqual(epoch_counter, 3)
self.assertEqual(batch_counter, 3)
self.assertEqual(tot_steps_counter, 2*reference_num_batches + 3)
mock_training_plan.training_args.return_value = {
'epochs': 2,
'batch_maxnum': None,
'num_updates': None,
}
iter_accountant = MiniBatchTrainingIterationsAccountant(mock_training_plan)
epoch_counter = 0
tot_steps_counter = 0
for epoch in iter_accountant.iterate_epochs():
epoch_counter += 1
batch_counter = 0
self.assertEqual(epoch_counter, iter_accountant.cur_epoch)
for batch in iter_accountant.iterate_batches():
batch_counter += 1
tot_steps_counter += 1
self.assertEqual(batch_counter, iter_accountant.cur_batch)
self.assertEqual(epoch_counter, 3)
self.assertEqual(batch_counter, 0)
self.assertEqual(tot_steps_counter, 2*reference_num_batches)
mock_training_plan.training_args.return_value = {
'epochs': 4,
'batch_maxnum': 3,
'num_updates': None,
}
iter_accountant = MiniBatchTrainingIterationsAccountant(mock_training_plan)
epoch_counter = 0
tot_steps_counter = 0
for epoch in iter_accountant.iterate_epochs():
epoch_counter += 1
batch_counter = 0
self.assertEqual(epoch_counter, iter_accountant.cur_epoch)
for batch in iter_accountant.iterate_batches():
batch_counter += 1
tot_steps_counter += 1
self.assertEqual(batch_counter, iter_accountant.cur_batch)
self.assertEqual(epoch_counter, 5)
self.assertEqual(batch_counter, 0)
self.assertEqual(tot_steps_counter, 3*4)
def test_mini_batch_training_iterations_accountaint_03_reporting(self):
mock_training_plan.training_args.return_value = {
'epochs': None,
'batch_maxnum': None,
'num_updates': 2*reference_num_batches + 3,
'log_interval': 1,
'batch_size': reference_batch_size
}
iter_accountant = MiniBatchTrainingIterationsAccountant(mock_training_plan)
tot_steps_counter = 0
for epoch in iter_accountant.iterate_epochs():
for batch in iter_accountant.iterate_batches():
tot_steps_counter += 1
iter_accountant.increment_sample_counters(reference_batch_size)
self.assertTrue(iter_accountant.should_log_this_batch())
num_samples, num_samples_max = iter_accountant.reporting_on_num_samples()
self.assertEqual(num_samples, tot_steps_counter*reference_batch_size)
self.assertEqual(num_samples_max, 2*reference_dataset_size + 3*reference_batch_size,
f'batches per epoch: {reference_num_batches}, batch size {reference_batch_size}')
num_iter, num_iter_max = iter_accountant.reporting_on_num_iter()
self.assertEqual(num_iter, tot_steps_counter)
self.assertEqual(num_iter_max, 2*reference_num_batches + 3,
f'batches per epoch: {reference_num_batches}, batch size {reference_batch_size}')
epoch_to_report = iter_accountant.reporting_on_epoch()
self.assertIsNone(epoch_to_report)
# testing the value of attribute (important since we need it for aggregator weights computation)
self.assertEqual(iter_accountant.num_samples_observed_in_total, tot_steps_counter * reference_batch_size)
mock_training_plan.training_args.return_value = {
'epochs': 2,
'batch_maxnum': None,
'num_updates': None,
'log_interval': 1,
'batch_size': reference_batch_size
}
iter_accountant = MiniBatchTrainingIterationsAccountant(mock_training_plan)
tot_steps_counter = 0
for epoch in iter_accountant.iterate_epochs():
for batch in iter_accountant.iterate_batches():
tot_steps_counter += 1
iter_accountant.increment_sample_counters(reference_batch_size)
self.assertTrue(iter_accountant.should_log_this_batch())
num_samples, num_samples_max = iter_accountant.reporting_on_num_samples()
self.assertEqual(num_samples, batch*reference_batch_size)
self.assertEqual(num_samples_max, reference_dataset_size)
num_iter, num_iter_max = iter_accountant.reporting_on_num_iter()
self.assertEqual(num_iter, batch)
self.assertEqual(num_iter_max, reference_num_batches)
epoch_to_report = iter_accountant.reporting_on_epoch()
self.assertEqual(epoch, epoch_to_report)
# testing the value of attribute (important since we need it for aggregator weights computation)
self.assertEqual(iter_accountant.num_samples_observed_in_total, tot_steps_counter * reference_batch_size)
if __name__ == '__main__': # pragma: no cover
unittest.main()