pytorch

1

2

3

4

5

6
from caffe2.python import core
7
import caffe2.python.hypothesis_test_util as hu
8
import caffe2.python.serialized_test.serialized_test_util as serial
9
import hypothesis.strategies as st
10
import numpy as np
11

12

13
class TestUnmaskOp(serial.SerializedTestCase):
14
    @serial.given(N=st.integers(min_value=2, max_value=20),
15
           dtype=st.sampled_from([
16
               np.bool_,
17
               np.int8,
18
               np.int16,
19
               np.int32,
20
               np.int64,
21
               np.uint8,
22
               np.uint16,
23
               np.float16,
24
               np.float32,
25
               np.float64]),
26
           **hu.gcs)
27
    def test(self, N, dtype, gc, dc):
28
        if dtype is np.bool_:
29
            all_value = np.random.choice(a=[True, False], size=N)
30
        else:
31
            all_value = (np.random.rand(N) * N).astype(dtype)
32

33
        M = np.random.randint(1, N)
34
        split = sorted(np.random.randint(1, N, size=M))
35
        indices = np.random.permutation(N)
36
        pieces = np.split(indices, split)
37

38
        def ref(*args, **kwargs):
39
            return (all_value,)
40

41
        inputs = []
42
        inputs_names = []
43
        for i, piece in enumerate(pieces):
44
            piece.sort()
45
            mask = np.zeros(N, dtype=np.bool_)
46
            mask[piece] = True
47
            values = all_value[piece]
48
            inputs.extend([mask, values])
49
            inputs_names.extend(["mask%d" % i, "value%d" % i])
50

51
        op = core.CreateOperator(
52
            'BooleanUnmask',
53
            inputs_names,
54
            'output')
55

56
        self.assertReferenceChecks(gc, op, inputs, ref)
57
        self.assertDeviceChecks(dc, op, inputs, [0])
58

59

60
if __name__ == "__main__":
61
    import unittest
62
    unittest.main()
63