1 : |
agomez |
1 |
#include "regressionline.h" |
2 : |
|
|
#include "mathfunc.h" |
3 : |
|
|
#include "errorhandler.h" |
4 : |
|
|
#include "gadget.h" |
5 : |
|
|
#include "global.h" |
6 : |
|
|
|
7 : |
|
|
// ********************************************************
|
8 : |
|
|
// Functions for Regression
|
9 : |
|
|
// ********************************************************
|
10 : |
|
|
Regression::Regression() {
|
11 : |
|
|
linetype = FREE;
|
12 : |
|
|
error = 1;
|
13 : |
|
|
useweights = 0;
|
14 : |
|
|
sse = a = b = 0.0;
|
15 : |
|
|
}
|
16 : |
|
|
|
17 : |
|
|
Regression::Regression(LineType ltype) {
|
18 : |
|
|
linetype = ltype;
|
19 : |
|
|
error = 0;
|
20 : |
|
|
useweights = 0;
|
21 : |
|
|
sse = a = b = 0.0;
|
22 : |
|
|
}
|
23 : |
|
|
|
24 : |
|
|
void Regression::setWeights(const DoubleVector& weights) {
|
25 : |
|
|
if (!useweights) {
|
26 : |
|
|
handle.logMessage(LOGWARN, "Warning in regression - unexpected use of weights");
|
27 : |
|
|
error = 1;
|
28 : |
|
|
return;
|
29 : |
|
|
}
|
30 : |
|
|
w = weights;
|
31 : |
|
|
}
|
32 : |
|
|
|
33 : |
|
|
double Regression::getSSE() {
|
34 : |
|
|
if (error)
|
35 : |
|
|
return verybig;
|
36 : |
|
|
return sse;
|
37 : |
|
|
}
|
38 : |
|
|
|
39 : |
|
|
void Regression::calcFit() {
|
40 : |
|
|
if (error)
|
41 : |
|
|
return;
|
42 : |
|
|
|
43 : |
|
|
switch (linetype) {
|
44 : |
|
|
case FREE:
|
45 : |
|
|
//need to calculate both the slope and the intercept
|
46 : |
|
|
this->calcSlopeIntercept();
|
47 : |
|
|
break;
|
48 : |
|
|
case FIXEDSLOPE:
|
49 : |
|
|
//only need to calculate the intercept
|
50 : |
|
|
this->calcIntercept();
|
51 : |
|
|
break;
|
52 : |
|
|
case FIXEDINTERCEPT:
|
53 : |
|
|
//only need to calculate the slope
|
54 : |
|
|
this->calcSlope();
|
55 : |
|
|
break;
|
56 : |
|
|
case FIXED:
|
57 : |
|
|
//nothing to be done here
|
58 : |
|
|
break;
|
59 : |
|
|
default:
|
60 : |
|
|
handle.logMessage(LOGWARN, "Warning in regression - unrecognised linetype", linetype);
|
61 : |
|
|
break;
|
62 : |
|
|
}
|
63 : |
|
|
|
64 : |
|
|
//finally calculate the SSE value
|
65 : |
|
|
if (useweights)
|
66 : |
|
|
this->calcSSEWeights();
|
67 : |
|
|
else
|
68 : |
|
|
this->calcSSE();
|
69 : |
|
|
}
|
70 : |
|
|
|
71 : |
|
|
void Regression::calcSSE() {
|
72 : |
|
|
if (error)
|
73 : |
|
|
return;
|
74 : |
|
|
|
75 : |
|
|
int i;
|
76 : |
|
|
double tmp;
|
77 : |
|
|
sse = 0.0;
|
78 : |
|
|
for (i = 0; i < x.Size(); i++) {
|
79 : |
|
|
tmp = y[i] - (a + b * x[i]);
|
80 : |
|
|
sse += tmp * tmp;
|
81 : |
|
|
}
|
82 : |
|
|
}
|
83 : |
|
|
|
84 : |
|
|
void Regression::calcSSEWeights() {
|
85 : |
|
|
if (error)
|
86 : |
|
|
return;
|
87 : |
|
|
|
88 : |
|
|
int i;
|
89 : |
|
|
double tmp;
|
90 : |
|
|
sse = 0.0;
|
91 : |
|
|
for (i = 0; i < x.Size(); i++) {
|
92 : |
|
|
tmp = y[i] - (a + b * x[i]);
|
93 : |
|
|
sse += w[i] * tmp * tmp;
|
94 : |
|
|
}
|
95 : |
|
|
}
|
96 : |
|
|
|
97 : |
|
|
void Regression::calcSlope() {
|
98 : |
|
|
if (error)
|
99 : |
|
|
return;
|
100 : |
|
|
|
101 : |
|
|
int i;
|
102 : |
|
|
double sumX, sumY;
|
103 : |
|
|
sumX = sumY = 0.0;
|
104 : |
|
|
for (i = 0; i < x.Size(); i++) {
|
105 : |
|
|
sumX += x[i];
|
106 : |
|
|
sumY += y[i];
|
107 : |
|
|
}
|
108 : |
|
|
if (isZero(sumX))
|
109 : |
|
|
b = 0.0;
|
110 : |
|
|
else
|
111 : |
|
|
b = (sumY - (a * x.Size())) / sumX;
|
112 : |
|
|
|
113 : |
|
|
//JMB - if there is a negative slope for the regression then things are going wrong
|
114 : |
|
|
if (b < 0.0) {
|
115 : |
|
|
handle.logMessage(LOGWARN, "Warning in regression - negative slope for regression line", b);
|
116 : |
|
|
error = 1;
|
117 : |
|
|
}
|
118 : |
|
|
}
|
119 : |
|
|
|
120 : |
|
|
void Regression::calcIntercept() {
|
121 : |
|
|
if (error)
|
122 : |
|
|
return;
|
123 : |
|
|
|
124 : |
|
|
int i;
|
125 : |
|
|
double sumX, sumY;
|
126 : |
|
|
sumX = sumY = 0.0;
|
127 : |
|
|
for (i = 0; i < x.Size(); i++) {
|
128 : |
|
|
sumX += x[i];
|
129 : |
|
|
sumY += y[i];
|
130 : |
|
|
}
|
131 : |
|
|
a = (sumY - (b * sumX)) / x.Size();
|
132 : |
|
|
}
|
133 : |
|
|
|
134 : |
|
|
void Regression::calcSlopeIntercept() {
|
135 : |
|
|
if (error)
|
136 : |
|
|
return;
|
137 : |
|
|
|
138 : |
|
|
int i;
|
139 : |
|
|
double sumX, sumY, nom, denom;
|
140 : |
|
|
sumX = sumY = nom = denom = 0.0;
|
141 : |
|
|
for (i = 0; i < x.Size(); i++) {
|
142 : |
|
|
sumX += x[i];
|
143 : |
|
|
sumY += y[i];
|
144 : |
|
|
}
|
145 : |
|
|
sumX /= x.Size();
|
146 : |
|
|
sumY /= y.Size();
|
147 : |
|
|
|
148 : |
|
|
for (i = 0; i < x.Size(); i++) {
|
149 : |
|
|
nom += (x[i] - sumX) * (y[i] - sumY);
|
150 : |
|
|
denom += (x[i] - sumX) * (x[i] - sumX);
|
151 : |
|
|
}
|
152 : |
|
|
|
153 : |
|
|
if (isZero(denom)) {
|
154 : |
|
|
b = 0.0;
|
155 : |
|
|
a = sumY;
|
156 : |
|
|
} else {
|
157 : |
|
|
b = nom / denom;
|
158 : |
|
|
a = sumY - (b * sumX);
|
159 : |
|
|
}
|
160 : |
|
|
|
161 : |
|
|
//JMB - if there is a negative slope for the regression then things are going wrong
|
162 : |
|
|
if (b < 0.0) {
|
163 : |
|
|
handle.logMessage(LOGWARN, "Warning in regression - negative slope for regression line", b);
|
164 : |
|
|
error = 1;
|
165 : |
|
|
}
|
166 : |
|
|
}
|
167 : |
|
|
|
168 : |
|
|
// ********************************************************
|
169 : |
|
|
// Functions for LinearRegression
|
170 : |
|
|
// ********************************************************
|
171 : |
|
|
LinearRegression::LinearRegression(LineType ltype) : Regression(ltype) {
|
172 : |
|
|
}
|
173 : |
|
|
|
174 : |
|
|
void LinearRegression::storeVectors(const DoubleVector& modData, const DoubleVector& obsData) {
|
175 : |
|
|
error = 0; //begin by cleaning up error status
|
176 : |
|
|
if ((modData.Size() != obsData.Size()) || (modData.Size() < 2)) {
|
177 : |
|
|
handle.logMessage(LOGWARN, "Warning in linear regression - invalid vector sizes");
|
178 : |
|
|
error = 1;
|
179 : |
|
|
return;
|
180 : |
|
|
}
|
181 : |
|
|
x = modData;
|
182 : |
|
|
y = obsData;
|
183 : |
|
|
}
|
184 : |
|
|
|
185 : |
|
|
// ********************************************************
|
186 : |
|
|
// Functions for LogLinearRegression
|
187 : |
|
|
// ********************************************************
|
188 : |
|
|
LogLinearRegression::LogLinearRegression(LineType ltype) : Regression(ltype) {
|
189 : |
|
|
}
|
190 : |
|
|
|
191 : |
|
|
void LogLinearRegression::storeVectors(const DoubleVector& modData, const DoubleVector& obsData) {
|
192 : |
|
|
error = 0; //begin by cleaning up error status
|
193 : |
|
|
if ((modData.Size() != obsData.Size()) || (modData.Size() < 2)) {
|
194 : |
|
|
handle.logMessage(LOGWARN, "Warning in log linear regression - invalid vector sizes");
|
195 : |
|
|
error = 1;
|
196 : |
|
|
return;
|
197 : |
|
|
}
|
198 : |
|
|
|
199 : |
|
|
x.Reset();
|
200 : |
|
|
x.resize(modData.Size(), 0.0);
|
201 : |
|
|
y.Reset();
|
202 : |
|
|
y.resize(obsData.Size(), 0.0);
|
203 : |
|
|
|
204 : |
|
|
int i, l = 0;
|
205 : |
|
|
for (i = 0; i < x.Size(); i++, l++) {
|
206 : |
|
|
if (isZero(modData[i]) && isZero(obsData[i])) {
|
207 : |
|
|
//omit the point (0.0, 0.0)
|
208 : |
|
|
x.Delete(l);
|
209 : |
|
|
y.Delete(l);
|
210 : |
|
|
l--;
|
211 : |
|
|
} else if ((modData[i] < verysmall) || (obsData[i] < verysmall)) {
|
212 : |
|
|
handle.logMessage(LOGWARN, "Warning in log linear regession - received invalid values");
|
213 : |
|
|
error = 1;
|
214 : |
|
|
return;
|
215 : |
|
|
} else {
|
216 : |
|
|
x[l] = log(modData[i]);
|
217 : |
|
|
y[l] = log(obsData[i]);
|
218 : |
|
|
}
|
219 : |
|
|
}
|
220 : |
|
|
}
|
221 : |
|
|
|
222 : |
|
|
// ********************************************************
|
223 : |
|
|
// Functions for WeightRegression
|
224 : |
|
|
// ********************************************************
|
225 : |
|
|
WeightRegression::WeightRegression(LineType ltype) : LinearRegression(ltype) {
|
226 : |
|
|
useweights = 1;
|
227 : |
|
|
}
|
228 : |
|
|
|
229 : |
|
|
void WeightRegression::storeVectors(const DoubleVector& modData, const DoubleVector& obsData) {
|
230 : |
|
|
LinearRegression::storeVectors(modData, obsData);
|
231 : |
|
|
if (x.Size() != w.Size()) {
|
232 : |
|
|
handle.logMessage(LOGWARN, "Warning in weight regression - invalid vector sizes");
|
233 : |
|
|
error = 1;
|
234 : |
|
|
}
|
235 : |
|
|
}
|
236 : |
|
|
|
237 : |
|
|
// ********************************************************
|
238 : |
|
|
// Functions for LogWeightRegression
|
239 : |
|
|
// ********************************************************
|
240 : |
|
|
LogWeightRegression::LogWeightRegression(LineType ltype) : LogLinearRegression(ltype) {
|
241 : |
|
|
useweights = 1;
|
242 : |
|
|
}
|
243 : |
|
|
|
244 : |
|
|
void LogWeightRegression::storeVectors(const DoubleVector& modData, const DoubleVector& obsData) {
|
245 : |
|
|
LogLinearRegression::storeVectors(modData, obsData);
|
246 : |
|
|
if (x.Size() != w.Size()) {
|
247 : |
|
|
handle.logMessage(LOGWARN, "Warning in log weight regression - invalid vector sizes");
|
248 : |
|
|
error = 1;
|
249 : |
|
|
}
|
250 : |
|
|
}
|