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
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
|
from caffe2.python import core, workspace
import caffe2.python.hypothesis_test_util as hu
import caffe2.python.serialized_test.serialized_test_util as serial
from hypothesis import given, settings
from hypothesis import strategies as st
import numpy as np
import time
class TestTensorPackOps(serial.SerializedTestCase):
def pack_segments_ref(self, return_presence_mask=False, max_length=None):
def pack_segments_ref(lengths, data, max_length=max_length):
arr = []
constant_values = 0
if data.dtype.char == 'S':
constant_values = ''
if max_length is None:
max_length = np.max(lengths)
start = 0
for idx in range(np.size(lengths)):
len = lengths[idx] if max_length >= lengths[idx] else max_length
chunk = data[start : start + len]
pad_length = max_length - len
# ((0, pad_length), (0, 0)) says add pad_length rows of padding
# below chunk and 0 rows of padding elsewhere
arr.append(
np.pad(
chunk, ((0, pad_length), (0, 0)),
mode=str("constant"),
constant_values=constant_values
)
)
start += lengths[idx]
result = [arr]
if return_presence_mask:
presence_arr = []
for length in lengths:
length = length if max_length >= length else max_length
pad_length = max_length - length
presence_arr.append(
np.pad(
np.ones((length), dtype=np.bool), ((0, pad_length)),
mode=str("constant")
)
)
result.append(presence_arr)
return result
return pack_segments_ref
@given(
num_seq=st.integers(10, 100),
cell_size=st.integers(1, 10),
max_length_buffer=st.integers(-5, 5),
**hu.gcs
)
@settings(deadline=None, max_examples=50)
def test_pack_with_max_length_ops(
self, num_seq, cell_size, max_length_buffer, gc, dc
):
# create data
lengths = np.arange(num_seq, dtype=np.int32) + 1
num_cell = np.sum(lengths)
data = np.zeros(num_cell * cell_size, dtype=np.float32)
left = np.cumsum(np.arange(num_seq) * cell_size)
right = np.cumsum(lengths * cell_size)
for i in range(num_seq):
data[left[i]:right[i]] = i + 1.0
data.resize(num_cell, cell_size)
print("\nnum seq:{}, num cell: {}, cell size:{}\n".format(
num_seq, num_cell, cell_size)
+ "=" * 60
)
# run test
max_length = num_seq + max_length_buffer
op = core.CreateOperator(
'PackSegments', ['l', 'd'], ['t'], max_length=max_length)
workspace.FeedBlob('l', lengths)
workspace.FeedBlob('d', data)
start = time.time()
self.assertReferenceChecks(
device_option=gc,
op=op,
inputs=[lengths, data, max_length],
reference=self.pack_segments_ref(max_length=max_length),
)
end = time.time()
print("{} used time: {}".format(gc, end - start).replace('\n', ' '))
with core.DeviceScope(gc):
workspace.FeedBlob('l', lengths)
workspace.FeedBlob('d', data)
workspace.RunOperatorOnce(core.CreateOperator(
'PackSegments',
['l', 'd'],
['t'],
max_length=max_length,
device_option=gc))
workspace.RunOperatorOnce(core.CreateOperator(
'UnpackSegments',
['l', 't'],
['newd'],
max_length=max_length,
device_option=gc))
assert(workspace.FetchBlob('t').shape[1] == max_length)
def _cal_unpacked_data(data):
if max_length >= num_seq:
return data
output = None
start = 0
for i, length in enumerate(lengths):
new_len = max_length if length > max_length else length
chunk = data[start: start + new_len]
if output is None:
output = chunk
else:
output = np.concatenate((output, chunk), axis=0)
start += length
return output
true_newd = _cal_unpacked_data(workspace.FetchBlob('d'))
assert((workspace.FetchBlob('newd') == true_newd).all())
@given(
num_seq=st.integers(10, 500),
cell_size=st.integers(1, 10),
**hu.gcs
)
@settings(deadline=10000)
def test_pack_ops(self, num_seq, cell_size, gc, dc):
# create data
lengths = np.arange(num_seq, dtype=np.int32) + 1
num_cell = np.sum(lengths)
data = np.zeros(num_cell * cell_size, dtype=np.float32)
left = np.cumsum(np.arange(num_seq) * cell_size)
right = np.cumsum(lengths * cell_size)
for i in range(num_seq):
data[left[i]:right[i]] = i + 1.0
data.resize(num_cell, cell_size)
print("\nnum seq:{}, num cell: {}, cell size:{}\n".format(
num_seq, num_cell, cell_size)
+ "=" * 60
)
# run test
op = core.CreateOperator(
'PackSegments', ['l', 'd'], ['t'])
workspace.FeedBlob('l', lengths)
workspace.FeedBlob('d', data)
start = time.time()
self.assertReferenceChecks(
device_option=gc,
op=op,
inputs=[lengths, data],
reference=self.pack_segments_ref(),
)
end = time.time()
print("{} used time: {}".format(gc, end - start).replace('\n', ' '))
with core.DeviceScope(gc):
workspace.FeedBlob('l', lengths)
workspace.FeedBlob('d', data)
workspace.RunOperatorOnce(core.CreateOperator(
'PackSegments',
['l', 'd'],
['t'],
device_option=gc))
workspace.RunOperatorOnce(core.CreateOperator(
'UnpackSegments',
['l', 't'],
['newd'],
device_option=gc))
assert((workspace.FetchBlob('newd') == workspace.FetchBlob('d')).all())
@given(
**hu.gcs_cpu_only
)
def test_pack_ops_str(self, gc, dc):
# GPU does not support string. Test CPU implementation only.
workspace.FeedBlob('l', np.array([1, 2, 3], dtype=np.int64))
strs = np.array([
["a", "a"],
["b", "b"],
["bb", "bb"],
["c", "c"],
["cc", "cc"],
["ccc", "ccc"]],
dtype='|S')
workspace.FeedBlob('d', strs)
workspace.RunOperatorOnce(core.CreateOperator(
'PackSegments',
['l', 'd'],
['t'],
device_option=gc))
workspace.RunOperatorOnce(core.CreateOperator(
'UnpackSegments',
['l', 't'],
['newd'],
device_option=gc))
assert((workspace.FetchBlob('newd') == workspace.FetchBlob('d')).all())
def test_pad_minf(self):
workspace.FeedBlob('l', np.array([1, 2, 3], dtype=np.int32))
workspace.FeedBlob(
'd',
np.array([
[1.0, 1.1],
[2.0, 2.1],
[2.2, 2.2],
[3.0, 3.1],
[3.2, 3.3],
[3.4, 3.5]],
dtype=np.float32))
workspace.RunOperatorOnce(core.CreateOperator(
'PackSegments', ['l', 'd'], ['t'], pad_minf=True))
workspace.RunOperatorOnce(core.CreateOperator(
'Exp', ['t'], ['r']
))
result = workspace.FetchBlob('t')
assert(result[0, -1, 0] < -1000.0)
# The whole point of padding with -inf is that when we exponentiate it
# then it should be zero.
exponentiated = workspace.FetchBlob('r')
assert(exponentiated[0, -1, 0] == 0.0)
def test_pad_no_minf(self):
workspace.FeedBlob('l', np.array([1, 2, 3], dtype=np.int32))
workspace.FeedBlob(
'd',
np.array([
[1.0, 1.1],
[2.0, 2.1],
[2.2, 2.2],
[3.0, 3.1],
[3.2, 3.3],
[3.4, 3.5]],
dtype=np.float32))
workspace.RunOperatorOnce(
core.CreateOperator(
'PackSegments', ['l', 'd'], ['t'], pad_minf=False),
)
result = workspace.FetchBlob('t')
assert(result[0, -1, 0] == 0.0)
workspace.FeedBlob(
'i',
np.array([
[1, 1],
[2, 2],
[2, 2],
[3, 3],
[3, 3],
[3, 3]],
dtype=np.int32))
workspace.RunOperatorOnce(
core.CreateOperator(
'PackSegments', ['l', 'i'], ['t2'], pad_minf=False),
)
result = workspace.FetchBlob('t2')
assert(result[0, -1, 0] == 0)
@given(**hu.gcs)
def test_presence_mask(self, gc, dc):
lengths = np.array([1, 2, 3], dtype=np.int32)
data = np.array(
[
[1.0, 1.0], [2.0, 2.0], [2.0, 2.0], [3.0, 3.0], [3.0, 3.0],
[3.0, 3.0]
],
dtype=np.float32
)
op = core.CreateOperator(
'PackSegments', ['l', 'd'], ['t', 'p'], return_presence_mask=True
)
workspace.FeedBlob('l', lengths)
workspace.FeedBlob('d', data)
inputs = [lengths, data]
self.assertReferenceChecks(
device_option=gc,
op=op,
inputs=inputs,
reference=self.pack_segments_ref(return_presence_mask=True),
)
op = core.CreateOperator(
'PackSegments', ['l', 'd'], ['t', 'p'], return_presence_mask=True
)
workspace.RunOperatorOnce(op)
output = workspace.FetchBlob('t')
expected_output_shape = (3, 3, 2)
self.assertEquals(output.shape, expected_output_shape)
presence_mask = workspace.FetchBlob('p')
expected_presence_mask = np.array(
[[True, False, False], [True, True, False], [True, True, True]],
dtype=np.bool
)
self.assertEqual(presence_mask.shape, expected_presence_mask.shape)
np.testing.assert_array_equal(presence_mask, expected_presence_mask)
def test_presence_mask_empty(self):
lengths = np.array([], dtype=np.int32)
data = np.array([], dtype=np.float32)
op = core.CreateOperator(
'PackSegments', ['l', 'd'], ['t', 'p'], return_presence_mask=True
)
workspace.FeedBlob('l', lengths)
workspace.FeedBlob('d', data)
workspace.RunOperatorOnce(op)
output = workspace.FetchBlob('p')
expected_output_shape = (0, 0)
self.assertEquals(output.shape, expected_output_shape)
@given(**hu.gcs_cpu_only)
@settings(deadline=10000)
def test_out_of_bounds(self, gc, dc):
# Copy pasted from test_pack_ops but with 3 changed to 4
lengths = np.array([1, 2, 4], dtype=np.int32)
data = np.array([
[1.0, 1.0],
[2.0, 2.0],
[2.0, 2.0],
[3.0, 3.0],
[3.0, 3.0],
[3.0, 3.0]], dtype=np.float32)
op = core.CreateOperator(
'PackSegments', ['l', 'd'], ['t'])
inputs = [lengths, data]
self.assertRunOpRaises(
device_option=gc,
op=op,
inputs=inputs,
exception=RuntimeError
)
@given(**hu.gcs_cpu_only)
@settings(deadline=10000)
def test_under_bounds(self, gc, dc):
# Copy pasted from test_pack_ops but with 3 changed to 2
lengths = np.array([1, 2, 2], dtype=np.int32)
data = np.array([
[1.0, 1.0],
[2.0, 2.0],
[2.0, 2.0],
[3.0, 3.0],
[3.0, 3.0],
[3.0, 3.0]], dtype=np.float32)
op = core.CreateOperator(
'PackSegments', ['l', 'd'], ['t'])
inputs = [lengths, data]
self.assertRunOpRaises(
device_option=gc,
op=op,
inputs=inputs,
exception=RuntimeError
)
if __name__ == "__main__":
import unittest
unittest.main()
|
