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
|
// -*- Mode: weights_++; tab-width: 2; -*-
// vi: set ts=2:
//
//
#include <BALL/QSAR/plsModel.h>
#include <Eigen/Dense>
namespace BALL
{
namespace QSAR
{
PLSModel::PLSModel(const QSARData& q) : LinearModel(q)
{
type_="PLS";
no_components_ = 10;
}
PLSModel::~PLSModel()
{
}
int PLSModel::getNoComponents()
{
return no_components_;
}
void PLSModel::setNoComponents(int no)
{
no_components_ = no;
}
void PLSModel::train()
{
if (descriptor_matrix_.cols() == 0)
{
throw Exception::InconsistentUsage(__FILE__, __LINE__, "Data must be read into the model before training!");
}
int cols = descriptor_matrix_.cols();
Eigen::MatrixXd X = descriptor_matrix_;
Eigen::MatrixXd P; // Matrix P saves all vectors p
// determine the number of components that are to be created.
// no_components_ contains the number of components desired by the user,
// but obviously we cannot calculate more PLS components than features
unsigned int components_to_create = no_components_;
if (cols < no_components_) components_to_create = cols;
U_.resize(X.rows(), components_to_create);
loadings_.resize(cols, components_to_create);
weights_.resize(Y_.cols(), components_to_create);
latent_variables_.resize(descriptor_matrix_.rows(), components_to_create);
P.resize(cols, components_to_create);
Eigen::VectorXd w;
Eigen::VectorXd t;
Eigen::VectorXd c;
Eigen::VectorXd u = Y_.col(0);
Eigen::VectorXd u_old;
for (unsigned int j = 0; j < components_to_create; j++)
{
for (int i = 0; i < 10000; i++)
{
w = X.transpose()*u / Statistics::scalarProduct(u);
w = w / Statistics::euclNorm(w);
t = X*w;
c = Y_.transpose()*t / Statistics::scalarProduct(t);
u_old = u;
u = Y_*c / Statistics::scalarProduct(c);
if (Statistics::euclDistance(u, u_old)/Statistics::euclNorm(u) > 0.0000001)
{
continue;
}
else // if u has converged
{
break;
}
}
Eigen::VectorXd p = X.transpose()*t / Statistics::scalarProduct(t);
X -= t * p.transpose();
U_.col(j) = u;
loadings_.col(j) = w;
weights_.col(j) = c;
P.col(j) = p;
latent_variables_.col(j) = t;
}
// try // p's are not orthogonal to each other, so that in rare cases P.t()*loadings_ is not invertible
// {
// loadings_ = loadings_*(P.t()*loadings_).i();
// }
// catch(BALL::Exception::MatrixIsSingular e)
// {
// Eigen::MatrixXd I; I.setToIdentity(P.cols());
// I *= 0.0001;
// loadings_ = loadings_*(P.t()*loadings_+I).i();
// }
training_result_ = loadings_*(P.transpose()*loadings_).jacobiSvd(Eigen::ComputeThinU | Eigen::ComputeThinV).solve(weights_.transpose());
calculateOffsets();
}
bool PLSModel::optimizeParameters(int k, int no_steps)
{
double best_q2 = 0;
int best_no = 1;
for (int i = 1; i <= no_steps && i <= (int)data->getNoDescriptors() && (descriptor_IDs_.empty() || i <= (int)descriptor_IDs_.size()); i++)
{
no_components_ = i;
validation->crossValidation(k);
if (validation->getQ2() > best_q2)
{
best_q2 = validation->getQ2();
best_no = i;
}
else if (validation->getQ2() < 0.75*best_q2)
{
break; // for speed-up
}
}
no_components_ = best_no;
validation->setQ2(best_q2);
return 1;
}
const Eigen::MatrixXd* PLSModel::getU()
{
return &U_;
}
void PLSModel::setParameters(vector<double>& v)
{
if (v.size() != 1)
{
String c = "Wrong number of model parameters! Needed: 1;";
c = c+" given: "+String(v.size());
throw Exception::ModelParameterError(__FILE__, __LINE__, c.c_str());
}
no_components_ = (int) v[0];
}
vector<double> PLSModel::getParameters() const
{
vector<double> v;
v.push_back(no_components_);
return v;
}
}
}
|
