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
|
// 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.operations.Softmax;
import ai.vespa.rankingexpression.importer.operations.Sum;
import com.yahoo.searchlib.rankingexpression.evaluation.TensorValue;
import ai.vespa.rankingexpression.importer.IntermediateGraph;
import ai.vespa.rankingexpression.importer.OrderedTensorType;
import ai.vespa.rankingexpression.importer.operations.Argument;
import ai.vespa.rankingexpression.importer.operations.ConcatV2;
import ai.vespa.rankingexpression.importer.operations.Const;
import ai.vespa.rankingexpression.importer.operations.Constant;
import ai.vespa.rankingexpression.importer.operations.ExpandDims;
import ai.vespa.rankingexpression.importer.operations.Identity;
import ai.vespa.rankingexpression.importer.operations.IntermediateOperation;
import ai.vespa.rankingexpression.importer.operations.Join;
import ai.vespa.rankingexpression.importer.operations.Map;
import ai.vespa.rankingexpression.importer.operations.MatMul;
import ai.vespa.rankingexpression.importer.operations.Mean;
import ai.vespa.rankingexpression.importer.operations.Merge;
import ai.vespa.rankingexpression.importer.operations.NoOp;
import ai.vespa.rankingexpression.importer.operations.PlaceholderWithDefault;
import ai.vespa.rankingexpression.importer.operations.Reshape;
import ai.vespa.rankingexpression.importer.operations.Select;
import ai.vespa.rankingexpression.importer.operations.Shape;
import ai.vespa.rankingexpression.importer.operations.Squeeze;
import ai.vespa.rankingexpression.importer.operations.Switch;
import com.yahoo.tensor.functions.ScalarFunctions;
import org.tensorflow.SavedModelBundle;
import org.tensorflow.Session;
import org.tensorflow.framework.GraphDef;
import org.tensorflow.framework.MetaGraphDef;
import org.tensorflow.framework.NodeDef;
import org.tensorflow.framework.SignatureDef;
import org.tensorflow.framework.TensorInfo;
import java.io.IOException;
import java.util.List;
import java.util.stream.Collectors;
/**
* Converts a TensorFlow graph to a Vespa IntermediateGraph which is the basis
* for generating Vespa ranking expressions.
*
* @author lesters
*/
class GraphImporter {
private static IntermediateOperation mapOperation(NodeDef node,
List<IntermediateOperation> inputs,
IntermediateGraph graph) {
String nodeName = node.getName();
String modelName = graph.name();
int nodePort = IntermediateOperation.indexPartOf(nodeName);
OrderedTensorType nodeType = TypeConverter.typeFrom(node);
AttributeConverter attributes = AttributeConverter.convert(node);
switch (node.getOp().toLowerCase()) {
// array ops
case "concatv2": return new ConcatV2(modelName, nodeName, inputs);
case "const": return new Const(modelName, nodeName, inputs, attributes, nodeType);
case "expanddims": return new ExpandDims(modelName, nodeName, inputs);
case "identity": return new Identity(modelName, nodeName, inputs);
case "placeholder": return new Argument(modelName, nodeName, nodeType);
case "placeholderwithdefault": return new PlaceholderWithDefault(modelName, nodeName, inputs);
case "reshape": return new Reshape(modelName, nodeName, inputs, attributes);
case "shape": return new Shape(modelName, nodeName, inputs);
case "squeeze": return new Squeeze(modelName, nodeName, inputs, attributes);
// control flow
case "merge": return new Merge(modelName, nodeName, inputs);
case "switch": return new Switch(modelName, nodeName, inputs, nodePort);
// math ops
case "abs": return new Map(modelName, nodeName, inputs, ScalarFunctions.abs());
case "acos": return new Map(modelName, nodeName, inputs, ScalarFunctions.acos());
case "add": return new Join(modelName, nodeName, inputs, ScalarFunctions.add());
case "add_n": return new Join(modelName, nodeName, inputs, ScalarFunctions.add());
case "asin": return new Map(modelName, nodeName, inputs, ScalarFunctions.asin());
case "atan": return new Map(modelName, nodeName, inputs, ScalarFunctions.atan());
case "ceil": return new Map(modelName, nodeName, inputs, ScalarFunctions.ceil());
case "cos": return new Map(modelName, nodeName, inputs, ScalarFunctions.cos());
case "div": return new Join(modelName, nodeName, inputs, ScalarFunctions.divide());
case "exp": return new Map(modelName, nodeName, inputs, ScalarFunctions.exp());
case "realdiv": return new Join(modelName, nodeName, inputs, ScalarFunctions.divide());
case "floor": return new Map(modelName, nodeName, inputs, ScalarFunctions.floor());
case "log": return new Map(modelName, nodeName, inputs, ScalarFunctions.log());
case "matmul": return new MatMul(modelName, nodeName, inputs);
case "maximum": return new Join(modelName, nodeName, inputs, ScalarFunctions.max());
case "mean": return new Mean(modelName, nodeName, inputs, attributes);
case "reducemean": return new Mean(modelName, nodeName, inputs, attributes);
case "mul": return new Join(modelName, nodeName, inputs, ScalarFunctions.multiply());
case "multiply": return new Join(modelName, nodeName, inputs, ScalarFunctions.multiply());
case "negate": return new Map(modelName, nodeName, inputs, ScalarFunctions.neg());
case "reciprocal": return new Map(modelName, nodeName, inputs, ScalarFunctions.reciprocal());
case "rsqrt": return new Map(modelName, nodeName, inputs, ScalarFunctions.rsqrt());
case "select": return new Select(modelName, nodeName, inputs);
case "where3": return new Select(modelName, nodeName, inputs);
case "sigmoid": return new Map(modelName, nodeName, inputs, ScalarFunctions.sigmoid());
case "sin": return new Map(modelName, nodeName, inputs, ScalarFunctions.sin());
case "squareddifference": return new Join(modelName, nodeName, inputs, ScalarFunctions.squareddifference());
case "sub": return new Join(modelName, nodeName, inputs, ScalarFunctions.subtract());
case "subtract": return new Join(modelName, nodeName, inputs, ScalarFunctions.subtract());
case "sum": return new Sum(modelName, nodeName, inputs, attributes);
case "square": return new Map(modelName, nodeName, inputs, ScalarFunctions.square());
case "sqrt": return new Map(modelName, nodeName, inputs, ScalarFunctions.sqrt());
case "tan": return new Map(modelName, nodeName, inputs, ScalarFunctions.tan());
case "tanh": return new Map(modelName, nodeName, inputs, ScalarFunctions.tanh());
// nn ops
case "biasadd": return new Join(modelName, nodeName, inputs, ScalarFunctions.add());
case "elu": return new Map(modelName, nodeName, inputs, ScalarFunctions.elu());
case "relu": return new Map(modelName, nodeName, inputs, ScalarFunctions.relu());
case "selu": return new Map(modelName, nodeName, inputs, ScalarFunctions.selu());
case "softmax": return new Softmax(modelName, nodeName, inputs, attributes);
// state ops
case "variable": return new Constant(modelName, nodeName, nodeType);
case "variablev2": return new Constant(modelName, nodeName, nodeType);
case "varhandleop": return new Constant(modelName, nodeName, nodeType);
case "readvariableop":return new Identity(modelName, nodeName, inputs);
// evaluation no-ops
case "stopgradient":return new Identity(modelName, nodeName, inputs);
case "noop": return new NoOp(modelName, nodeName, inputs);
}
IntermediateOperation op = new NoOp(modelName, node.getName(), inputs);
op.warning("Operation '" + node.getOp() + "' is currently not implemented");
return op;
}
static IntermediateGraph importGraph(String modelName, SavedModelBundle bundle) throws IOException {
MetaGraphDef tfGraph = MetaGraphDef.parseFrom(bundle.metaGraphDef());
IntermediateGraph intermediateGraph = new IntermediateGraph(modelName);
importSignatures(tfGraph, intermediateGraph);
importOperations(tfGraph, intermediateGraph, bundle);
verifyOutputTypes(tfGraph, intermediateGraph);
return intermediateGraph;
}
private static void importSignatures(MetaGraphDef tfGraph, IntermediateGraph intermediateGraph) {
for (java.util.Map.Entry<String, SignatureDef> signatureEntry : tfGraph.getSignatureDefMap().entrySet()) {
String signatureName = signatureEntry.getKey();
java.util.Map<String, TensorInfo> inputInfoMap = signatureEntry.getValue().getInputsMap();
for (java.util.Map.Entry<String, TensorInfo> input : inputInfoMap.entrySet()) {
String inputName = input.getKey();
String nodeName = input.getValue().getName();
intermediateGraph.inputs(signatureName).put(inputName, IntermediateOperation.namePartOf(nodeName));
}
java.util.Map<String, TensorInfo> outputInfoMap = signatureEntry.getValue().getOutputsMap();
for (java.util.Map.Entry<String, TensorInfo> output : outputInfoMap.entrySet()) {
String outputName = output.getKey();
String nodeName = output.getValue().getName();
intermediateGraph.outputs(signatureName).put(outputName, IntermediateOperation.namePartOf(nodeName));
}
}
}
private static void importOperations(MetaGraphDef tfGraph,
IntermediateGraph intermediateGraph,
SavedModelBundle bundle) {
for (String signatureName : intermediateGraph.signatures()) {
for (String outputName : intermediateGraph.outputs(signatureName).values()) {
importOperation(outputName, tfGraph.getGraphDef(), intermediateGraph, bundle);
}
}
}
private static IntermediateOperation importOperation(String nodeName,
GraphDef tfGraph,
IntermediateGraph intermediateGraph,
SavedModelBundle bundle) {
if (intermediateGraph.alreadyImported(nodeName)) {
return intermediateGraph.get(nodeName);
}
NodeDef node = getTensorFlowNodeFromGraph(IntermediateOperation.namePartOf(nodeName), tfGraph);
List<IntermediateOperation> inputs = importOperationInputs(node, tfGraph, intermediateGraph, bundle);
IntermediateOperation operation = mapOperation(node, inputs, intermediateGraph);
intermediateGraph.put(nodeName, operation);
List<IntermediateOperation> controlInputs = importControlInputs(node, tfGraph, intermediateGraph, bundle);
if (controlInputs.size() > 0) {
operation.setControlInputs(controlInputs);
}
if (operation.isConstant()) {
operation.setConstantValueFunction(
type -> new TensorValue(TensorConverter.toVespaTensor(readVariable(nodeName, bundle), type)));
}
return operation;
}
private static List<IntermediateOperation> importOperationInputs(NodeDef node,
GraphDef tfGraph,
IntermediateGraph intermediateGraph,
SavedModelBundle bundle) {
return node.getInputList().stream()
.filter(name -> ! isControlDependency(name))
.map(nodeName -> importOperation(nodeName, tfGraph, intermediateGraph, bundle))
.collect(Collectors.toList());
}
private static List<IntermediateOperation> importControlInputs(NodeDef node,
GraphDef tfGraph,
IntermediateGraph intermediateGraph,
SavedModelBundle bundle) {
return node.getInputList().stream()
.filter(nodeName -> isControlDependency(nodeName))
.map(nodeName -> importOperation(nodeName, tfGraph, intermediateGraph, bundle))
.collect(Collectors.toList());
}
private static boolean isControlDependency(String name) {
return name.startsWith("^");
}
private static NodeDef getTensorFlowNodeFromGraph(String name, GraphDef tfGraph) {
for (NodeDef node : tfGraph.getNodeList()) {
if (node.getName().equals(name)) {
return node;
}
}
throw new IllegalArgumentException("Could not find node '" + name + "'");
}
static org.tensorflow.Tensor<?> readVariable(String name, SavedModelBundle bundle) {
Session.Runner fetched = bundle.session().runner().fetch(name);
List<org.tensorflow.Tensor<?>> importedTensors = fetched.run();
if (importedTensors.size() != 1)
throw new IllegalStateException("Expected 1 tensor from fetching " + name +
", but got " + importedTensors.size());
return importedTensors.get(0);
}
private static void verifyOutputTypes(MetaGraphDef tfGraph, IntermediateGraph intermediateGraph) {
for (String signatureName : intermediateGraph.signatures()) {
for (String outputName : intermediateGraph.outputs(signatureName).values()) {
IntermediateOperation operation = intermediateGraph.get(outputName);
NodeDef node = getTensorFlowNodeFromGraph(IntermediateOperation.namePartOf(operation.name()), tfGraph.getGraphDef());
OrderedTensorType type = operation.type().orElseThrow(
() -> new IllegalArgumentException("Output of '" + outputName + "' has no type."));
TypeConverter.verifyType(node, type);
}
}
}
}
|
