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
|
// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
package com.yahoo.tensor.functions;
import com.yahoo.tensor.evaluation.EvaluationContext;
import com.yahoo.tensor.evaluation.Name;
import com.yahoo.tensor.evaluation.TypeContext;
import com.yahoo.tensor.Tensor;
import com.yahoo.tensor.TensorType;
import com.yahoo.tensor.TensorType.Dimension;
import java.util.List;
import java.util.Objects;
/**
* Convenience for euclidean distance between vectors.
* euclidean_distance(a, b, mydim) == sqrt(sum(pow(a-b, 2), mydim))
* @author arnej
*/
public class EuclideanDistance<NAMETYPE extends Name> extends TensorFunction<NAMETYPE> {
private final TensorFunction<NAMETYPE> arg1;
private final TensorFunction<NAMETYPE> arg2;
private final String dimension;
public EuclideanDistance(TensorFunction<NAMETYPE> argument1,
TensorFunction<NAMETYPE> argument2,
String dimension)
{
this.arg1 = argument1;
this.arg2 = argument2;
this.dimension = dimension;
}
@Override
public List<TensorFunction<NAMETYPE>> arguments() { return List.of(arg1, arg2); }
@Override
public TensorFunction<NAMETYPE> withArguments(List<TensorFunction<NAMETYPE>> arguments) {
if ( arguments.size() != 2)
throw new IllegalArgumentException("EuclideanDistance must have 2 arguments, got " + arguments.size());
return new EuclideanDistance<>(arguments.get(0), arguments.get(1), dimension);
}
@Override
public TensorType type(TypeContext<NAMETYPE> context) {
TensorType t1 = arg1.toPrimitive().type(context);
TensorType t2 = arg2.toPrimitive().type(context);
var d1 = t1.dimension(dimension);
var d2 = t2.dimension(dimension);
if (d1.isEmpty() || d2.isEmpty()
|| d1.get().type() != Dimension.Type.indexedBound
|| d2.get().type() != Dimension.Type.indexedBound
|| ! d1.get().size().equals(d2.get().size()))
{
throw new IllegalArgumentException("euclidean_distance expects both arguments to have the '"
+ dimension + "' dimension with same size, but input types were "
+ t1 + " and " + t2);
}
// Finds the type this produces by first converting it to a primitive function
return toPrimitive().type(context);
}
/** Evaluates this by first converting it to a primitive function */
@Override
public Tensor evaluate(EvaluationContext<NAMETYPE> context) {
return toPrimitive().evaluate(context);
}
@Override
public PrimitiveTensorFunction<NAMETYPE> toPrimitive() {
TensorFunction<NAMETYPE> primitive1 = arg1.toPrimitive();
TensorFunction<NAMETYPE> primitive2 = arg2.toPrimitive();
// this should match the C++ optimized "l2_distance"
var diffs = new Join<>(primitive1, primitive2, ScalarFunctions.subtract());
var squaredDiffs = new Map<>(diffs, ScalarFunctions.square());
var sumOfSquares = new Reduce<>(squaredDiffs, Reduce.Aggregator.sum, dimension);
return new Map<>(sumOfSquares, ScalarFunctions.sqrt());
}
@Override
public String toString(ToStringContext<NAMETYPE> context) {
return "euclidean_distance(" + arg1.toString(context) + ", " + arg2.toString(context) + ", " + dimension + ")";
}
@Override
public int hashCode() { return Objects.hash("euclidean_distance", arg1, arg2, dimension); }
}
|
