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
|
CLASS:: GaussClass
summary:: Gaussian classifier
categories:: UGens
DESCRIPTION::
A Gaussian classifier, which classifies an input vector as belonging to one of the gaussian distributions defined in a specially-formatted Buffer.
The Buffer should be single-channel. Its exact format is specified towards the bottom of this file. If you have the MathLib quark installed then you can use the convenience function GaussClass.classesToFloatArray() to create a FloatArray suitable for loading to a Buffer.
CLASSMETHODS::
METHOD:: kr
argument:: in
input signal, a multichannel signal specifying a co-ordinate in the space (i.e. a vector).
argument:: bufnum
the buffer in which the shapes and weights of the gaussian components are specified.
argument:: gate
the classifier is only active when this is greater than 0 (otherwise, previous value is held constant). Its default value is 1.
EXAMPLES::
The following example creates two-dimensional data with three classes:
code::
(
~classes = [
[ // First class's mean, covariance, weight:
[2, 8], [[1, 0], [0, 1]], 0.3
],[ // Second class's mean, covariance, weight:
[8, 2], [[1, 0], [0, 1]], 0.3
],[ // Third class's mean, covariance, weight:
[8, 8], [[0.75, 0.5], [0.5, 0.75]], 0.4
]
];
~serialised = GaussClass.classesToFloatArray(~classes);
)
// Now let's use it:
s.boot;
b = Buffer.loadCollection(s, ~serialised);
(
x = {
var rate = 20, input, result, gate;
//input = {LFNoise2.kr(rate).range(0, 10)}.dup(2); // Our "query point" wanders around in space
input = [MouseX.kr(0, 10), MouseY.kr(0, 10)]; // Or you can wander yourself using the mouse
gate = Impulse.kr(rate);
result = GaussClass.kr(input, b, gate);
input.poll(gate, "input");
result.poll(gate, "result");
Out.ar(0, SinOsc.ar(result.linexp(0, ~classes.size-1, 440, 880), 0, 0.1).dup); // sonify
}.play(s)
)
x.free;
::
________________________________
THE FORMAT OF THE BUFFER:
The Buffer should be single-channel and hold data in the following order, once for each class:
- N floats: the mean vector;
- N*N floats: the inverse of the covariance matrix; and
- 1 float: the weight of the component divided by the square root of the determinant of the covariance matrix.
N is the dimensionality of the data space. The length of the Buffer is therefore C * (N*N + N + 1). GaussClass.kr will determine the number of classes from the size of the Buffer.
|
