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
|
// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
package com.yahoo.schema.expressiontransforms;
import com.yahoo.schema.FeatureNames;
import com.yahoo.searchlib.rankingexpression.RankingExpression;
import com.yahoo.searchlib.rankingexpression.Reference;
import com.yahoo.searchlib.rankingexpression.parser.ParseException;
import com.yahoo.searchlib.rankingexpression.rule.CompositeNode;
import com.yahoo.searchlib.rankingexpression.rule.ConstantNode;
import com.yahoo.searchlib.rankingexpression.rule.ExpressionNode;
import com.yahoo.searchlib.rankingexpression.rule.ReferenceNode;
import com.yahoo.searchlib.rankingexpression.rule.TensorFunctionNode;
import com.yahoo.searchlib.rankingexpression.transform.ExpressionTransformer;
import com.yahoo.tensor.functions.Generate;
import java.io.StringReader;
import java.util.Collection;
import java.util.HashSet;
import java.util.Set;
import java.util.logging.Logger;
/**
* Analyzes expression to figure out what inputs it needs
*
* @author arnej
*/
public class InputRecorder extends ExpressionTransformer<InputRecorderContext> {
private static final Logger log = Logger.getLogger(InputRecorder.class.getName());
private final Set<String> neededInputs;
private final Set<String> handled = new HashSet<>();
private final Set<String> availableNormalizers = new HashSet<>();
private final Set<String> usedNormalizers = new HashSet<>();
public InputRecorder(Set<String> target) {
this.neededInputs = target;
}
public void process(RankingExpression expression, RankProfileTransformContext context) {
process(expression.getRoot(), context);
}
public void process(ExpressionNode node, RankProfileTransformContext context) {
transform(node, new InputRecorderContext(context));
}
public void alreadyMatchFeatures(Collection<String> matchFeatures) {
for (String mf : matchFeatures) {
handled.add(mf);
}
}
public void addKnownNormalizers(Collection<String> names) {
for (String name : names) {
availableNormalizers.add(name);
}
}
public Set<String> normalizersUsed() { return this.usedNormalizers; }
@Override
public ExpressionNode transform(ExpressionNode node, InputRecorderContext context) {
if (node instanceof ReferenceNode r) {
handle(r, context);
return node;
}
if (node instanceof TensorFunctionNode t) {
var f = t.function();
if (f instanceof Generate) {
var childContext = new InputRecorderContext(context);
var tt = f.type(context.types());
// expects only indexed dimensions, should we check?
for (var dim : tt.dimensions()) {
childContext.localVariables().add(dim.name());
}
return transformChildren(t, childContext);
}
node = t.withTransformedExpressions(expr -> transform(expr, context));
}
if (node instanceof CompositeNode c)
return transformChildren(c, context);
if (node instanceof ConstantNode) {
return node;
}
throw new IllegalArgumentException("Cannot handle node type: "+ node + " [" + node.getClass() + "]");
}
private void handle(ReferenceNode feature, InputRecorderContext context) {
Reference ref = feature.reference();
String name = ref.name();
var args = ref.arguments();
boolean simpleFunctionOrIdentifier = (args.size() == 0) && (ref.output() == null);
if (simpleFunctionOrIdentifier && context.localVariables().contains(name)) {
return;
}
if (simpleFunctionOrIdentifier && availableNormalizers.contains(name)) {
usedNormalizers.add(name);
return;
}
if (ref.isSimpleRankingExpressionWrapper()) {
name = ref.simpleArgument().get();
simpleFunctionOrIdentifier = true;
}
if (simpleFunctionOrIdentifier) {
if (handled.contains(name)) {
return;
}
var f = context.rankProfile().getFunctions().get(name);
if (f != null && f.function().arguments().size() == 0) {
transform(f.function().getBody().getRoot(), context);
handled.add(name);
return;
}
neededInputs.add(feature.toString());
return;
}
if (FeatureNames.isSimpleFeature(ref)) {
if (FeatureNames.isAttributeFeature(ref)) {
neededInputs.add(feature.toString());
return;
}
if (FeatureNames.isQueryFeature(ref)) {
// we should be able to get it from the query
return;
}
if (FeatureNames.isConstantFeature(ref)) {
var allConstants = context.rankProfile().constants();
if (allConstants.containsKey(ref)) {
// assumes we have the constant available during evaluation without any more wiring
return;
}
throw new IllegalArgumentException("unknown constant: " + feature);
}
}
if ("onnx".equals(name)) {
if (args.size() < 1) {
throw new IllegalArgumentException("expected name of ONNX model as argument: " + feature);
}
var arg = args.expressions().get(0);
var models = context.rankProfile().onnxModels();
var model = models.get(arg.toString());
if (model == null) {
var tmp = OnnxModelTransformer.transformFeature(feature, context.rankProfile());
if (tmp instanceof ReferenceNode newRefNode) {
args = newRefNode.getArguments();
arg = args.expressions().get(0);
model = models.get(arg.toString());
}
}
if (model == null) {
throw new IllegalArgumentException("missing onnx model: " + arg);
}
model.getInputMap().forEach((__, onnxInput) -> {
var reader = new StringReader(onnxInput);
try {
var asExpression = new RankingExpression(reader);
transform(asExpression.getRoot(), context);
} catch (ParseException e) {
throw new IllegalArgumentException("illegal onnx input '" + onnxInput + "': " + e.getMessage());
}
});
return;
}
neededInputs.add(feature.toString());
}
}
|
