blob: 08c0564ed8a956e0917a61780a9e7dfa7d29db93 (
plain) (
blame)
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
|
// Copyright 2018 Yahoo Holdings. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
package ai.vespa.rankingexpression.importer.tensorflow;
import ai.vespa.rankingexpression.importer.OrderedTensorType;
import com.yahoo.tensor.TensorType;
import org.tensorflow.framework.AttrValue;
import org.tensorflow.framework.DataType;
import org.tensorflow.framework.NodeDef;
import org.tensorflow.framework.TensorShapeProto;
/**
* Converts and verifies TensorFlow tensor types into Vespa tensor types.
*
* @author lesters
*/
class TypeConverter {
static void verifyType(NodeDef node, OrderedTensorType type) {
TensorShapeProto shape = tensorFlowShape(node);
if (shape != null) {
if (shape.getDimCount() != type.rank()) {
throw new IllegalArgumentException("TensorFlow shape of '" + node.getName() + "' " +
"does not match Vespa shape");
}
for (int tensorFlowIndex = 0; tensorFlowIndex < type.dimensions().size(); ++tensorFlowIndex) {
int vespaIndex = type.dimensionMap(tensorFlowIndex);
TensorShapeProto.Dim tensorFlowDimension = shape.getDim(tensorFlowIndex);
TensorType.Dimension vespaDimension = type.type().dimensions().get(vespaIndex);
if (tensorFlowDimension.getSize() != vespaDimension.size().orElse(-1L)) {
throw new IllegalArgumentException("TensorFlow dimensions of '" + node.getName() + "' " +
"does not match Vespa dimensions");
}
}
}
}
static OrderedTensorType typeFrom(NodeDef node) {
String dimensionPrefix = "d"; // standard naming convention: d0, d1, ...
TensorShapeProto shape = tensorFlowShape(node);
OrderedTensorType.Builder builder = new OrderedTensorType.Builder(toValueType(tensorFlowValueType(node)));
for (int i = 0; i < shape.getDimCount(); ++ i) {
String dimensionName = dimensionPrefix + i;
TensorShapeProto.Dim tensorFlowDimension = shape.getDim(i);
if (tensorFlowDimension.getSize() >= 0) {
builder.add(TensorType.Dimension.indexed(dimensionName, tensorFlowDimension.getSize()));
} else {
builder.add(TensorType.Dimension.indexed(dimensionName));
}
}
return builder.build();
}
static TensorType typeFrom(org.tensorflow.Tensor<?> tfTensor, String dimensionPrefix) {
TensorType.Builder b = new TensorType.Builder(toValueType(tfTensor.dataType()));
int dimensionIndex = 0;
for (long dimensionSize : tfTensor.shape()) {
if (dimensionSize == 0) dimensionSize = 1; // TensorFlow ...
b.indexed(dimensionPrefix + (dimensionIndex++), dimensionSize);
}
return b.build();
}
private static TensorShapeProto tensorFlowShape(NodeDef node) {
AttrValue attrValueList = node.getAttrMap().get("_output_shapes");
if (attrValueList == null)
throw new IllegalArgumentException("_output_shapes attribute of '" + node.getName() + "' " +
"does not exist");
if (attrValueList.getValueCase() != AttrValue.ValueCase.LIST)
throw new IllegalArgumentException("_output_shapes attribute of '" + node.getName() + "' " +
"is not of expected type");
return attrValueList.getList().getShape(0); // support multiple outputs?
}
private static DataType tensorFlowValueType(NodeDef node) {
AttrValue attrValueList = node.getAttrMap().get("dtypes");
if (attrValueList == null)
return DataType.DT_DOUBLE; // default. This will usually (always?) be used. TODO: How can we do better?
if (attrValueList.getValueCase() != AttrValue.ValueCase.LIST)
return DataType.DT_DOUBLE; // default
return attrValueList.getList().getType(0); // support multiple outputs?
}
private static TensorType.Value toValueType(DataType dataType) {
switch (dataType) {
case DT_FLOAT: return TensorType.Value.DOUBLE;
case DT_DOUBLE: return TensorType.Value.DOUBLE;
// Imperfect conversion, for now:
case DT_BOOL: return TensorType.Value.DOUBLE;
case DT_BFLOAT16: return TensorType.Value.DOUBLE;
case DT_HALF: return TensorType.Value.DOUBLE;
case DT_INT8: return TensorType.Value.DOUBLE;
case DT_INT16: return TensorType.Value.DOUBLE;
case DT_INT32: return TensorType.Value.DOUBLE;
case DT_INT64: return TensorType.Value.DOUBLE;
case DT_UINT8: return TensorType.Value.DOUBLE;
case DT_UINT16: return TensorType.Value.DOUBLE;
case DT_UINT32: return TensorType.Value.DOUBLE;
case DT_UINT64: return TensorType.Value.DOUBLE;
default: throw new IllegalArgumentException("A TensorFlow tensor with data type " + dataType +
" cannot be converted to a Vespa tensor type");
}
}
private static TensorType.Value toValueType(org.tensorflow.DataType dataType) {
switch (dataType) {
case FLOAT: return TensorType.Value.DOUBLE;
case DOUBLE: return TensorType.Value.DOUBLE;
// Imperfect conversion, for now:
case BOOL: return TensorType.Value.DOUBLE;
case INT32: return TensorType.Value.DOUBLE;
case UINT8: return TensorType.Value.DOUBLE;
case INT64: return TensorType.Value.DOUBLE;
default: throw new IllegalArgumentException("A TensorFlow tensor with data type " + dataType +
" cannot be converted to a Vespa tensor type");
}
}
}
|
