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
|
// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
package com.yahoo.tensor.serialization;
import com.yahoo.io.GrowableByteBuffer;
import com.yahoo.tensor.MixedTensor;
import com.yahoo.tensor.Tensor;
import com.yahoo.tensor.TensorAddress;
import com.yahoo.tensor.TensorType;
import org.junit.Test;
import java.util.Arrays;
import java.util.Optional;
import static org.junit.Assert.assertEquals;
import static org.junit.Assert.fail;
/**
* Tests for the sparse binary format.
*
* @author geirst
*/
public class SparseBinaryFormatTestCase {
@Test
public void testSerialization() {
assertSerialization("tensor(x{}):{}");
assertSerialization("tensor(x{}):{{x:0}:2.0}");
assertSerialization("tensor(dimX{},dimY{}):{{dimX:labelA,dimY:labelB}:2.0,{dimY:labelC,dimX:labelD}:3.0}");
assertSerialization("tensor(x{},y{}):{{x:0,y:1}:2.0}");
assertSerialization("tensor(x{},y{}):{{x:0,y:1}:2.0,{x:1,y:4}:3.0}");
assertSerialization("tensor(x{},y{},z{}):{{y:0,x:0,z:3}:2.0}");
assertSerialization("tensor(x{},y{},z{}):{{y:0,x:0,z:3}:2.0,{y:1,x:0,z:6}:3.0}");
}
@Test
public void testSerializationFormatIsDecidedByTensorTypeNotImplementationType() {
Tensor sparse = Tensor.Builder.of(TensorType.fromSpec("tensor(x{})"))
.cell(TensorAddress.ofLabels("key1"), 9.1).build();
Tensor sparseAsMixed = MixedTensor.Builder.of(TensorType.fromSpec("tensor(x{})"))
.cell(TensorAddress.ofLabels("key1"), 9.1).build();
byte[] sparseEncoded = TypedBinaryFormat.encode(sparse);
byte[] sparseAsMixedEncoded = TypedBinaryFormat.encode(sparseAsMixed);
assertEquals(Arrays.toString(sparseEncoded), Arrays.toString(sparseAsMixedEncoded));
}
@Test
public void testSerializationToSeparateType() {
try {
assertSerialization(Tensor.from("tensor(x{},y{}):{{x:0,y:0}:2.0}"), TensorType.fromSpec("tensor(x{})"));
fail("Expected exception");
}
catch (IllegalArgumentException expected) {
assertEquals("Type/instance mismatch: A tensor of type tensor(x{},y{}) cannot be assigned to type tensor(x{})", expected.getMessage());
}
}
@Test
public void requireThatSerializationFormatDoesNotChange() {
byte[] encodedTensor = new byte[] {1, // binary format type
2, // num dimensions
2, (byte)'x', (byte)'y', 1, (byte)'z', // dimensions
2, // num cells,
2, (byte)'a', (byte)'b', 1, (byte)'e', 64, 0, 0, 0, 0, 0, 0, 0, // cell 0
2, (byte)'c', (byte)'d', 1, (byte)'e', 64, 8, 0, 0, 0, 0, 0, 0}; // cell 1
Tensor tensor = Tensor.from("tensor(xy{},z{}):{{xy:ab,z:e}:2.0,{xy:cd,z:e}:3.0}");
assertEquals(Arrays.toString(encodedTensor), Arrays.toString(TypedBinaryFormat.encode(tensor)));
}
@Test
public void requireThatFloatSerializationFormatDoesNotChange() {
byte[] encodedTensor = new byte[] {
5, // binary format type
1, // float type
2, // num dimensions
2, (byte)'x', (byte)'y', 1, (byte)'z', // dimensions
2, // num cells,
2, (byte)'a', (byte)'b', 1, (byte)'e', 64, 0, 0, 0, // cell 0
2, (byte)'c', (byte)'d', 1, (byte)'e', 64, 64, 0, 0}; // cell 1
Tensor tensor = Tensor.from("tensor<float>(xy{},z{}):{{xy:ab,z:e}:2.0,{xy:cd,z:e}:3.0}");
assertEquals(Arrays.toString(encodedTensor), Arrays.toString(TypedBinaryFormat.encode(tensor)));
}
@Test
public void requireThatBFloat16SerializationFormatDoesNotChange() {
byte[] encodedTensor = new byte[] {
5, // binary format type
2, // bfloat16 type
2, // num dimensions
2, (byte)'x', (byte)'y', 1, (byte)'z', // dimensions
2, // num cells,
2, (byte)'a', (byte)'b', 1, (byte)'e', 64, 0, // cell 0
2, (byte)'c', (byte)'d', 1, (byte)'e', 64, 64}; // cell 1
Tensor tensor = Tensor.from("tensor<bfloat16>(xy{},z{}):{{xy:ab,z:e}:2.0,{xy:cd,z:e}:3.0}");
assertEquals(Arrays.toString(encodedTensor), Arrays.toString(TypedBinaryFormat.encode(tensor)));
}
@Test
public void requireThatInt8SerializationFormatDoesNotChange() {
byte[] encodedTensor = new byte[] {
5, // binary format type
3, // int8 type
2, // num dimensions
2, (byte)'x', (byte)'y', 1, (byte)'z', // dimensions
2, // num cells,
2, (byte)'a', (byte)'b', 1, (byte)'e', 2, // cell 0
2, (byte)'c', (byte)'d', 1, (byte)'e', 3}; // cell 1
Tensor tensor = Tensor.from("tensor<int8>(xy{},z{}):{{xy:ab,z:e}:2.0,{xy:cd,z:e}:3.0}");
assertEquals(Arrays.toString(encodedTensor), Arrays.toString(TypedBinaryFormat.encode(tensor)));
}
@Test
public void testSerializationOfDifferentValueTypes() {
assertSerialization("tensor<double>(x{},y{}):{{x:0,y:0}:2.0, {x:0,y:1}:3.0, {x:1,y:0}:4.0, {x:1,y:1}:5.0}");
assertSerialization("tensor<float>(x{},y{}):{{x:0,y:0}:2.0, {x:0,y:1}:3.0, {x:1,y:0}:4.0, {x:1,y:1}:5.0}");
assertSerialization("tensor<bfloat16>(x{},y{}):{{x:0,y:0}:2.0, {x:0,y:1}:3.0, {x:1,y:0}:4.0, {x:1,y:1}:5.0}");
assertSerialization("tensor<int8>(x{},y{}):{{x:0,y:0}:2, {x:0,y:1}:3, {x:1,y:0}:4, {x:1,y:1}:5}");
}
private void assertSerialization(String tensorString) {
assertSerialization(Tensor.from(tensorString));
}
private void assertSerialization(Tensor tensor) {
assertSerialization(tensor, tensor.type());
}
private void assertSerialization(Tensor tensor, TensorType expectedType) {
byte[] encodedTensor = TypedBinaryFormat.encode(tensor);
Tensor decodedTensor = TypedBinaryFormat.decode(Optional.of(expectedType),
GrowableByteBuffer.wrap(encodedTensor));
assertEquals(tensor, decodedTensor);
}
}
|
