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
|
// Copyright 2016 Yahoo Inc. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
package com.yahoo.tensor.serialization;
import com.google.common.annotations.Beta;
import com.yahoo.io.GrowableByteBuffer;
import com.yahoo.tensor.MappedTensor;
import com.yahoo.tensor.Tensor;
import com.yahoo.tensor.TensorAddress;
import com.yahoo.tensor.TensorType;
import com.yahoo.text.Utf8;
import java.util.*;
/**
* Implementation of a sparse binary format for a tensor on the form:
*
* Sorted dimensions = num_dimensions [dimension_str_len dimension_str_bytes]*
* Cells = num_cells [label_1_str_len label_1_str_bytes ... label_N_str_len label_N_str_bytes cell_value]*
*
* Note that the dimensions are sorted and the tensor address labels are given in the same sorted order.
* Unspecified labels are encoded as the empty string "".
*
* @author geirst
*/
@Beta
class SparseBinaryFormat implements BinaryFormat {
@Override
public void encode(GrowableByteBuffer buffer, Tensor tensor) {
encodeDimensions(buffer, tensor.type().dimensions());
encodeCells(buffer, tensor);
}
private void encodeDimensions(GrowableByteBuffer buffer, List<TensorType.Dimension> sortedDimensions) {
buffer.putInt1_4Bytes(sortedDimensions.size());
for (TensorType.Dimension dimension : sortedDimensions) {
buffer.putUtf8String(dimension.name());
}
}
private void encodeCells(GrowableByteBuffer buffer, Tensor tensor) {
buffer.putInt1_4Bytes(tensor.size());
for (Iterator<Tensor.Cell> i = tensor.cellIterator(); i.hasNext(); ) {
Map.Entry<TensorAddress, Double> cell = i.next();
encodeAddress(buffer, cell.getKey());
buffer.putDouble(cell.getValue());
}
}
private void encodeAddress(GrowableByteBuffer buffer, TensorAddress address) {
for (int i = 0; i < address.size(); i++)
buffer.putUtf8String(address.label(i));
}
@Override
public Tensor decode(TensorType type, GrowableByteBuffer buffer) {
consumeAndValidateDimensions(type, buffer);
Tensor.Builder builder = Tensor.Builder.of(type);
decodeCells(buffer, builder, type);
return builder.build();
}
private void consumeAndValidateDimensions(TensorType type, GrowableByteBuffer buffer) {
int dimensionCount = buffer.getInt1_4Bytes();
if (type.dimensions().size() != dimensionCount)
throw new IllegalArgumentException("Type/instance mismatch: Instance has " + dimensionCount +
" dimensions but type is " + type);
for (int i = 0; i < dimensionCount; ++i) {
TensorType.Dimension expectedDimension = type.dimensions().get(i);
String encodedName = buffer.getUtf8String();
if ( ! expectedDimension.name().equals(encodedName))
throw new IllegalArgumentException("Type/instance mismatch: Instance has '" + encodedName +
"' as dimension " + i + " but type is " + type);
}
}
private void decodeCells(GrowableByteBuffer buffer, Tensor.Builder builder, TensorType type) {
int numCells = buffer.getInt1_4Bytes();
for (int i = 0; i < numCells; ++i) {
Tensor.Builder.CellBuilder cellBuilder = builder.cell();
decodeAddress(buffer, cellBuilder, type);
cellBuilder.value(buffer.getDouble());
}
}
private void decodeAddress(GrowableByteBuffer buffer, Tensor.Builder.CellBuilder builder, TensorType type) {
for (TensorType.Dimension dimension : type.dimensions()) {
String label = buffer.getUtf8String();
if ( ! label.isEmpty()) {
builder.label(dimension.name(), label);
}
}
}
}
|