1 : |
agomez |
1 |
#include "catchdistribution.h" |
2 : |
|
|
#include "readfunc.h" |
3 : |
|
|
#include "readword.h" |
4 : |
|
|
#include "readaggregation.h" |
5 : |
|
|
#include "errorhandler.h" |
6 : |
|
|
#include "areatime.h" |
7 : |
|
|
#include "fleet.h" |
8 : |
|
|
#include "stock.h" |
9 : |
|
|
#include "multinomial.h" |
10 : |
|
|
#include "mathfunc.h" |
11 : |
|
|
#include "stockprey.h" |
12 : |
|
|
#include "ludecomposition.h" |
13 : |
|
|
#include "gadget.h" |
14 : |
|
|
#include "global.h" |
15 : |
|
|
|
16 : |
|
|
CatchDistribution::CatchDistribution(CommentStream& infile, const AreaClass* const Area,
|
17 : |
|
|
const TimeClass* const TimeInfo, Keeper* const keeper, double weight, const char* name)
|
18 : |
|
|
: Likelihood(CATCHDISTRIBUTIONLIKELIHOOD, weight, name), alptr(0) {
|
19 : |
|
|
|
20 : |
|
|
int i, j;
|
21 : |
|
|
char text[MaxStrLength];
|
22 : |
|
|
strncpy(text, "", MaxStrLength);
|
23 : |
|
|
int numarea = 0, numage = 0, numlen = 0;
|
24 : |
|
|
|
25 : |
|
|
char datafilename[MaxStrLength];
|
26 : |
|
|
char aggfilename[MaxStrLength];
|
27 : |
|
|
strncpy(datafilename, "", MaxStrLength);
|
28 : |
|
|
strncpy(aggfilename, "", MaxStrLength);
|
29 : |
|
|
ifstream datafile;
|
30 : |
|
|
CommentStream subdata(datafile);
|
31 : |
|
|
|
32 : |
|
|
timeindex = 0;
|
33 : |
|
|
yearly = 0;
|
34 : |
|
|
functionname = new char[MaxStrLength];
|
35 : |
|
|
strncpy(functionname, "", MaxStrLength);
|
36 : |
|
|
|
37 : |
|
|
readWordAndValue(infile, "datafile", datafilename);
|
38 : |
|
|
readWordAndValue(infile, "function", functionname);
|
39 : |
|
|
|
40 : |
|
|
functionnumber = 0;
|
41 : |
|
|
if (strcasecmp(functionname, "multinomial") == 0) {
|
42 : |
|
|
MN = Multinomial();
|
43 : |
|
|
functionnumber = 1;
|
44 : |
|
|
} else if (strcasecmp(functionname, "pearson") == 0) {
|
45 : |
|
|
functionnumber = 2;
|
46 : |
|
|
} else if (strcasecmp(functionname, "gamma") == 0) {
|
47 : |
|
|
functionnumber = 3;
|
48 : |
|
|
} else if (strcasecmp(functionname, "sumofsquares") == 0) {
|
49 : |
|
|
functionnumber = 4;
|
50 : |
|
|
} else if (strcasecmp(functionname, "mvn") == 0) {
|
51 : |
|
|
functionnumber = 5;
|
52 : |
|
|
|
53 : |
|
|
readWordAndVariable(infile, "lag", lag);
|
54 : |
|
|
readWordAndVariable(infile, "sigma", sigma);
|
55 : |
|
|
sigma.Inform(keeper);
|
56 : |
|
|
|
57 : |
|
|
params.resize(lag, keeper);
|
58 : |
|
|
for (i = 0; i < lag; i++)
|
59 : |
|
|
readWordAndVariable(infile, "param", params[i]);
|
60 : |
|
|
params.Inform(keeper);
|
61 : |
|
|
|
62 : |
|
|
} else if (strcasecmp(functionname, "mvlogistic") == 0) {
|
63 : |
|
|
functionnumber = 6;
|
64 : |
|
|
|
65 : |
|
|
readWordAndVariable(infile, "sigma", sigma);
|
66 : |
|
|
sigma.Inform(keeper);
|
67 : |
|
|
|
68 : |
|
|
} else if (strcasecmp(functionname, "log") == 0) {
|
69 : |
|
|
//JMB moved the logcatch function to here instead of it being a seperate class
|
70 : |
|
|
functionnumber = 7;
|
71 : |
|
|
|
72 : |
|
|
} else if (strcasecmp(functionname, "stratified") == 0) {
|
73 : |
|
|
//JMB experimental version of the sum of squares function for stratified samples
|
74 : |
|
|
functionnumber = 8;
|
75 : |
|
|
|
76 : |
|
|
} else
|
77 : |
|
|
handle.logFileMessage(LOGFAIL, "\nError in catchdistribution - unrecognised function", functionname);
|
78 : |
|
|
|
79 : |
|
|
infile >> ws;
|
80 : |
|
|
char c = infile.peek();
|
81 : |
|
|
if ((c == 'a') || (c == 'A')) {
|
82 : |
|
|
//we have found either aggregationlevel or areaaggfile ...
|
83 : |
|
|
streampos pos = infile.tellg();
|
84 : |
|
|
|
85 : |
|
|
infile >> text >> ws;
|
86 : |
|
|
if ((strcasecmp(text, "aggregation_level") == 0) || (strcasecmp(text, "aggregationlevel") == 0))
|
87 : |
|
|
infile >> yearly >> ws;
|
88 : |
|
|
else if (strcasecmp(text, "areaaggfile") == 0)
|
89 : |
|
|
infile.seekg(pos);
|
90 : |
|
|
else
|
91 : |
|
|
handle.logFileUnexpected(LOGFAIL, "areaaggfile", text);
|
92 : |
|
|
|
93 : |
|
|
//JMB - peek at the next char
|
94 : |
|
|
c = infile.peek();
|
95 : |
|
|
|
96 : |
|
|
if (yearly != 0 && yearly != 1)
|
97 : |
|
|
handle.logFileMessage(LOGFAIL, "\nError in catchdistribution - aggregationlevel must be 0 or 1");
|
98 : |
|
|
}
|
99 : |
|
|
|
100 : |
|
|
//JMB - changed to make the reading of overconsumption optional
|
101 : |
|
|
if ((c == 'o') || (c == 'O')) {
|
102 : |
|
|
readWordAndVariable(infile, "overconsumption", overconsumption);
|
103 : |
|
|
infile >> ws;
|
104 : |
|
|
c = infile.peek();
|
105 : |
|
|
} else
|
106 : |
|
|
overconsumption = 0;
|
107 : |
|
|
|
108 : |
|
|
if (overconsumption != 0 && overconsumption != 1)
|
109 : |
|
|
handle.logFileMessage(LOGFAIL, "\nError in catchdistribution - overconsumption must be 0 or 1");
|
110 : |
|
|
|
111 : |
|
|
//JMB - changed to make the reading of minimum probability optional
|
112 : |
|
|
if ((c == 'm') || (c == 'M'))
|
113 : |
|
|
readWordAndVariable(infile, "minimumprobability", epsilon);
|
114 : |
|
|
else if ((c == 'e') || (c == 'E'))
|
115 : |
|
|
readWordAndVariable(infile, "epsilon", epsilon);
|
116 : |
|
|
else
|
117 : |
|
|
epsilon = 10.0;
|
118 : |
|
|
|
119 : |
|
|
if (epsilon < verysmall) {
|
120 : |
|
|
handle.logFileMessage(LOGWARN, "epsilon should be a positive integer - set to default value 10");
|
121 : |
|
|
epsilon = 10.0;
|
122 : |
|
|
}
|
123 : |
|
|
|
124 : |
|
|
//read in area aggregation from file
|
125 : |
|
|
readWordAndValue(infile, "areaaggfile", aggfilename);
|
126 : |
|
|
datafile.open(aggfilename, ios::in);
|
127 : |
|
|
handle.checkIfFailure(datafile, aggfilename);
|
128 : |
|
|
handle.Open(aggfilename);
|
129 : |
|
|
numarea = readAggregation(subdata, areas, areaindex);
|
130 : |
|
|
handle.Close();
|
131 : |
|
|
datafile.close();
|
132 : |
|
|
datafile.clear();
|
133 : |
|
|
|
134 : |
|
|
//read in age aggregation from file
|
135 : |
|
|
readWordAndValue(infile, "ageaggfile", aggfilename);
|
136 : |
|
|
datafile.open(aggfilename, ios::in);
|
137 : |
|
|
handle.checkIfFailure(datafile, aggfilename);
|
138 : |
|
|
handle.Open(aggfilename);
|
139 : |
|
|
numage = readAggregation(subdata, ages, ageindex);
|
140 : |
|
|
handle.Close();
|
141 : |
|
|
datafile.close();
|
142 : |
|
|
datafile.clear();
|
143 : |
|
|
|
144 : |
|
|
//read in length aggregation from file
|
145 : |
|
|
readWordAndValue(infile, "lenaggfile", aggfilename);
|
146 : |
|
|
datafile.open(aggfilename, ios::in);
|
147 : |
|
|
handle.checkIfFailure(datafile, aggfilename);
|
148 : |
|
|
handle.Open(aggfilename);
|
149 : |
|
|
numlen = readLengthAggregation(subdata, lengths, lenindex);
|
150 : |
|
|
handle.Close();
|
151 : |
|
|
datafile.close();
|
152 : |
|
|
datafile.clear();
|
153 : |
|
|
|
154 : |
|
|
LgrpDiv = new LengthGroupDivision(lengths);
|
155 : |
|
|
if (LgrpDiv->Error())
|
156 : |
|
|
handle.logMessage(LOGFAIL, "Error in catchdistribution - failed to create length group");
|
157 : |
|
|
|
158 : |
|
|
//Must change from outer areas to inner areas.
|
159 : |
|
|
for (i = 0; i < areas.Nrow(); i++)
|
160 : |
|
|
for (j = 0; j < areas.Ncol(i); j++)
|
161 : |
|
|
areas[i][j] = Area->getInnerArea(areas[i][j]);
|
162 : |
|
|
|
163 : |
|
|
//read in the fleetnames
|
164 : |
|
|
i = 0;
|
165 : |
|
|
infile >> text >> ws;
|
166 : |
|
|
if (strcasecmp(text, "fleetnames") != 0)
|
167 : |
|
|
handle.logFileUnexpected(LOGFAIL, "fleetnames", text);
|
168 : |
|
|
infile >> text >> ws;
|
169 : |
|
|
while (!infile.eof() && (strcasecmp(text, "stocknames") != 0)) {
|
170 : |
|
|
fleetnames.resize(new char[strlen(text) + 1]);
|
171 : |
|
|
strcpy(fleetnames[i++], text);
|
172 : |
|
|
infile >> text >> ws;
|
173 : |
|
|
}
|
174 : |
|
|
if (fleetnames.Size() == 0)
|
175 : |
|
|
handle.logFileMessage(LOGFAIL, "\nError in catchdistribution - failed to read fleets");
|
176 : |
|
|
handle.logMessage(LOGMESSAGE, "Read fleet data - number of fleets", fleetnames.Size());
|
177 : |
|
|
|
178 : |
|
|
//read in the stocknames
|
179 : |
|
|
i = 0;
|
180 : |
|
|
if (strcasecmp(text, "stocknames") != 0)
|
181 : |
|
|
handle.logFileUnexpected(LOGFAIL, "stocknames", text);
|
182 : |
|
|
infile >> text;
|
183 : |
|
|
while (!infile.eof() && (strcasecmp(text, "[component]") != 0)) {
|
184 : |
|
|
infile >> ws;
|
185 : |
|
|
stocknames.resize(new char[strlen(text) + 1]);
|
186 : |
|
|
strcpy(stocknames[i++], text);
|
187 : |
|
|
infile >> text;
|
188 : |
|
|
}
|
189 : |
|
|
if (stocknames.Size() == 0)
|
190 : |
|
|
handle.logFileMessage(LOGFAIL, "\nError in catchdistribution - failed to read stocks");
|
191 : |
|
|
handle.logMessage(LOGMESSAGE, "Read stock data - number of stocks", stocknames.Size());
|
192 : |
|
|
|
193 : |
|
|
//We have now read in all the data from the main likelihood file
|
194 : |
|
|
//But we have to read in the statistics data from datafilename
|
195 : |
|
|
datafile.open(datafilename, ios::in);
|
196 : |
|
|
handle.checkIfFailure(datafile, datafilename);
|
197 : |
|
|
handle.Open(datafilename);
|
198 : |
|
|
readDistributionData(subdata, TimeInfo, numarea, numage, numlen);
|
199 : |
|
|
handle.Close();
|
200 : |
|
|
datafile.close();
|
201 : |
|
|
datafile.clear();
|
202 : |
|
|
|
203 : |
|
|
switch (functionnumber) {
|
204 : |
|
|
case 2:
|
205 : |
|
|
case 3:
|
206 : |
|
|
case 4:
|
207 : |
|
|
case 7:
|
208 : |
|
|
case 8:
|
209 : |
|
|
for (i = 0; i < numarea; i++) {
|
210 : |
|
|
modelYearData.resize(new DoubleMatrix(numage, numlen, 0.0));
|
211 : |
|
|
obsYearData.resize(new DoubleMatrix(numage, numlen, 0.0));
|
212 : |
|
|
}
|
213 : |
|
|
break;
|
214 : |
|
|
case 1:
|
215 : |
|
|
case 5:
|
216 : |
|
|
case 6:
|
217 : |
|
|
if (yearly)
|
218 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - yearly aggregation is ignored for function", functionname);
|
219 : |
|
|
yearly = 0;
|
220 : |
|
|
break;
|
221 : |
|
|
default:
|
222 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - unrecognised function", functionname);
|
223 : |
|
|
break;
|
224 : |
|
|
}
|
225 : |
|
|
}
|
226 : |
|
|
|
227 : |
|
|
void CatchDistribution::readDistributionData(CommentStream& infile,
|
228 : |
|
|
const TimeClass* TimeInfo, int numarea, int numage, int numlen) {
|
229 : |
|
|
|
230 : |
|
|
int i, year, step;
|
231 : |
|
|
double tmpnumber;
|
232 : |
|
|
char tmparea[MaxStrLength], tmpage[MaxStrLength], tmplen[MaxStrLength];
|
233 : |
|
|
strncpy(tmparea, "", MaxStrLength);
|
234 : |
|
|
strncpy(tmpage, "", MaxStrLength);
|
235 : |
|
|
strncpy(tmplen, "", MaxStrLength);
|
236 : |
|
|
int keepdata, timeid, ageid, areaid, lenid, count, reject;
|
237 : |
|
|
|
238 : |
|
|
//Check the number of columns in the inputfile
|
239 : |
|
|
infile >> ws;
|
240 : |
|
|
if (countColumns(infile) != 6)
|
241 : |
|
|
handle.logFileMessage(LOGFAIL, "wrong number of columns in inputfile - should be 6");
|
242 : |
|
|
|
243 : |
|
|
year = step = count = reject = 0;
|
244 : |
|
|
while (!infile.eof()) {
|
245 : |
|
|
keepdata = 1;
|
246 : |
|
|
infile >> year >> step >> tmparea >> tmpage >> tmplen >> tmpnumber >> ws;
|
247 : |
|
|
|
248 : |
|
|
//crude check to see if something has gone wrong and avoid infinite loops
|
249 : |
|
|
if (strlen(tmparea) == 0)
|
250 : |
|
|
handle.logFileMessage(LOGFAIL, "failed to read data from file");
|
251 : |
|
|
|
252 : |
|
|
//if tmparea is in areaindex find areaid, else dont keep the data
|
253 : |
|
|
areaid = -1;
|
254 : |
|
|
for (i = 0; i < areaindex.Size(); i++)
|
255 : |
|
|
if (strcasecmp(areaindex[i], tmparea) == 0)
|
256 : |
|
|
areaid = i;
|
257 : |
|
|
|
258 : |
|
|
if (areaid == -1)
|
259 : |
|
|
keepdata = 0;
|
260 : |
|
|
|
261 : |
|
|
//if tmpage is in ageindex find ageid, else dont keep the data
|
262 : |
|
|
ageid = -1;
|
263 : |
|
|
for (i = 0; i < ageindex.Size(); i++)
|
264 : |
|
|
if (strcasecmp(ageindex[i], tmpage) == 0)
|
265 : |
|
|
ageid = i;
|
266 : |
|
|
|
267 : |
|
|
if (ageid == -1)
|
268 : |
|
|
keepdata = 0;
|
269 : |
|
|
|
270 : |
|
|
//if tmplen is in lenindex find lenid, else dont keep the data
|
271 : |
|
|
lenid = -1;
|
272 : |
|
|
for (i = 0; i < lenindex.Size(); i++)
|
273 : |
|
|
if (strcasecmp(lenindex[i], tmplen) == 0)
|
274 : |
|
|
lenid = i;
|
275 : |
|
|
|
276 : |
|
|
if (lenid == -1)
|
277 : |
|
|
keepdata = 0;
|
278 : |
|
|
|
279 : |
|
|
//check if the year and step are in the simulation
|
280 : |
|
|
timeid = -1;
|
281 : |
|
|
if ((TimeInfo->isWithinPeriod(year, step)) && (keepdata == 1)) {
|
282 : |
|
|
//if this is a new timestep, resize to store the data
|
283 : |
|
|
for (i = 0; i < Years.Size(); i++)
|
284 : |
|
|
if ((Years[i] == year) && (Steps[i] == step))
|
285 : |
|
|
timeid = i;
|
286 : |
|
|
|
287 : |
|
|
if (timeid == -1) {
|
288 : |
|
|
Years.resize(1, year);
|
289 : |
|
|
Steps.resize(1, step);
|
290 : |
|
|
timeid = (Years.Size() - 1);
|
291 : |
|
|
|
292 : |
|
|
obsDistribution.resize();
|
293 : |
|
|
modelDistribution.resize();
|
294 : |
|
|
likelihoodValues.AddRows(1, numarea, 0.0);
|
295 : |
|
|
for (i = 0; i < numarea; i++) {
|
296 : |
|
|
obsDistribution[timeid].resize(new DoubleMatrix(numage, numlen, 0.0));
|
297 : |
|
|
modelDistribution[timeid].resize(new DoubleMatrix(numage, numlen, 0.0));
|
298 : |
|
|
}
|
299 : |
|
|
}
|
300 : |
|
|
|
301 : |
|
|
} else
|
302 : |
|
|
keepdata = 0;
|
303 : |
|
|
|
304 : |
|
|
if (keepdata == 1) {
|
305 : |
|
|
//distribution data is required, so store it
|
306 : |
|
|
count++;
|
307 : |
|
|
(*obsDistribution[timeid][areaid])[ageid][lenid] = tmpnumber;
|
308 : |
|
|
} else
|
309 : |
|
|
reject++; //count number of rejected data points read from file
|
310 : |
|
|
}
|
311 : |
|
|
|
312 : |
|
|
AAT.addActions(Years, Steps, TimeInfo);
|
313 : |
|
|
if (count == 0)
|
314 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - found no data in the data file for", this->getName());
|
315 : |
|
|
if (reject != 0)
|
316 : |
|
|
handle.logMessage(LOGMESSAGE, "Discarded invalid catchdistribution data - number of invalid entries", reject);
|
317 : |
|
|
handle.logMessage(LOGMESSAGE, "Read catchdistribution data file - number of entries", count);
|
318 : |
|
|
}
|
319 : |
|
|
|
320 : |
|
|
CatchDistribution::~CatchDistribution() {
|
321 : |
|
|
int i, j;
|
322 : |
|
|
for (i = 0; i < stocknames.Size(); i++)
|
323 : |
|
|
delete[] stocknames[i];
|
324 : |
|
|
for (i = 0; i < fleetnames.Size(); i++)
|
325 : |
|
|
delete[] fleetnames[i];
|
326 : |
|
|
for (i = 0; i < areaindex.Size(); i++)
|
327 : |
|
|
delete[] areaindex[i];
|
328 : |
|
|
for (i = 0; i < ageindex.Size(); i++)
|
329 : |
|
|
delete[] ageindex[i];
|
330 : |
|
|
for (i = 0; i < lenindex.Size(); i++)
|
331 : |
|
|
delete[] lenindex[i];
|
332 : |
|
|
for (i = 0; i < obsDistribution.Nrow(); i++)
|
333 : |
|
|
for (j = 0; j < obsDistribution.Ncol(i); j++) {
|
334 : |
|
|
delete obsDistribution[i][j];
|
335 : |
|
|
delete modelDistribution[i][j];
|
336 : |
|
|
}
|
337 : |
|
|
for (i = 0; i < modelYearData.Size(); i++) {
|
338 : |
|
|
delete modelYearData[i];
|
339 : |
|
|
delete obsYearData[i];
|
340 : |
|
|
}
|
341 : |
|
|
delete aggregator;
|
342 : |
|
|
delete LgrpDiv;
|
343 : |
|
|
delete[] functionname;
|
344 : |
|
|
}
|
345 : |
|
|
|
346 : |
|
|
void CatchDistribution::Reset(const Keeper* const keeper) {
|
347 : |
|
|
Likelihood::Reset(keeper);
|
348 : |
|
|
if (isZero(weight))
|
349 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - zero weight for", this->getName());
|
350 : |
|
|
|
351 : |
|
|
int i, j;
|
352 : |
|
|
for (i = 0; i < modelDistribution.Nrow(); i++)
|
353 : |
|
|
for (j = 0; j < modelDistribution.Ncol(i); j++)
|
354 : |
|
|
(*modelDistribution[i][j]).setToZero();
|
355 : |
|
|
if (yearly)
|
356 : |
|
|
for (i = 0; i < modelYearData.Size(); i++) {
|
357 : |
|
|
(*modelYearData[i]).setToZero();
|
358 : |
|
|
(*obsYearData[i]).setToZero();
|
359 : |
|
|
}
|
360 : |
|
|
|
361 : |
|
|
switch (functionnumber) {
|
362 : |
|
|
case 2:
|
363 : |
|
|
case 3:
|
364 : |
|
|
case 4:
|
365 : |
|
|
case 6:
|
366 : |
|
|
case 7:
|
367 : |
|
|
case 8:
|
368 : |
|
|
break;
|
369 : |
|
|
case 1:
|
370 : |
|
|
MN.setValue(epsilon);
|
371 : |
|
|
break;
|
372 : |
|
|
case 5:
|
373 : |
|
|
illegal = 0;
|
374 : |
|
|
this->calcCorrelation();
|
375 : |
|
|
if ((illegal) || (LU.isIllegal()))
|
376 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - multivariate normal out of bounds");
|
377 : |
|
|
break;
|
378 : |
|
|
default:
|
379 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - unrecognised function", functionname);
|
380 : |
|
|
break;
|
381 : |
|
|
}
|
382 : |
|
|
|
383 : |
|
|
if (handle.getLogLevel() >= LOGMESSAGE)
|
384 : |
|
|
handle.logMessage(LOGMESSAGE, "Reset catchdistribution component", this->getName());
|
385 : |
|
|
}
|
386 : |
|
|
|
387 : |
|
|
void CatchDistribution::Print(ofstream& outfile) const {
|
388 : |
|
|
|
389 : |
|
|
int i;
|
390 : |
|
|
outfile << "\nCatch Distribution " << this->getName() << " - likelihood value " << likelihood
|
391 : |
|
|
<< "\n\tFunction " << functionname << "\n\tStock names:";
|
392 : |
|
|
for (i = 0; i < stocknames.Size(); i++)
|
393 : |
|
|
outfile << sep << stocknames[i];
|
394 : |
|
|
outfile << "\n\tFleet names:";
|
395 : |
|
|
for (i = 0; i < fleetnames.Size(); i++)
|
396 : |
|
|
outfile << sep << fleetnames[i];
|
397 : |
|
|
outfile << endl;
|
398 : |
|
|
|
399 : |
|
|
switch (functionnumber) {
|
400 : |
|
|
case 1:
|
401 : |
|
|
case 2:
|
402 : |
|
|
case 3:
|
403 : |
|
|
case 4:
|
404 : |
|
|
case 7:
|
405 : |
|
|
case 8:
|
406 : |
|
|
break;
|
407 : |
|
|
case 5:
|
408 : |
|
|
outfile << "\tMultivariate normal distribution parameters: sigma " << sigma;
|
409 : |
|
|
for (i = 0; i < lag; i++)
|
410 : |
|
|
outfile << " param" << i + 1 << " " << params[i];
|
411 : |
|
|
outfile << endl;
|
412 : |
|
|
break;
|
413 : |
|
|
case 6:
|
414 : |
|
|
outfile << "\tMultivariate logistic distribution parameter: sigma " << sigma << endl;
|
415 : |
|
|
break;
|
416 : |
|
|
default:
|
417 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - unrecognised function", functionname);
|
418 : |
|
|
break;
|
419 : |
|
|
}
|
420 : |
|
|
|
421 : |
|
|
aggregator->Print(outfile);
|
422 : |
|
|
outfile.flush();
|
423 : |
|
|
}
|
424 : |
|
|
|
425 : |
|
|
void CatchDistribution::printLikelihood(ofstream& outfile, const TimeClass* const TimeInfo) {
|
426 : |
|
|
|
427 : |
|
|
if (!AAT.atCurrentTime(TimeInfo))
|
428 : |
|
|
return;
|
429 : |
|
|
|
430 : |
|
|
int i, area, age, len;
|
431 : |
|
|
timeindex = -1;
|
432 : |
|
|
for (i = 0; i < Years.Size(); i++)
|
433 : |
|
|
if ((Years[i] == TimeInfo->getYear()) && (Steps[i] == TimeInfo->getStep()))
|
434 : |
|
|
timeindex = i;
|
435 : |
|
|
if (timeindex == -1)
|
436 : |
|
|
handle.logMessage(LOGFAIL, "Error in catchdistribution - invalid timestep");
|
437 : |
|
|
|
438 : |
|
|
for (area = 0; area < modelDistribution.Ncol(timeindex); area++) {
|
439 : |
|
|
for (age = 0; age < modelDistribution[timeindex][area]->Nrow(); age++) {
|
440 : |
|
|
for (len = 0; len < modelDistribution[timeindex][area]->Ncol(age); len++) {
|
441 : |
|
|
outfile << setw(lowwidth) << Years[timeindex] << sep << setw(lowwidth)
|
442 : |
|
|
<< Steps[timeindex] << sep << setw(printwidth) << areaindex[area] << sep
|
443 : |
|
|
<< setw(printwidth) << ageindex[age] << sep << setw(printwidth)
|
444 : |
|
|
<< lenindex[len] << sep << setprecision(largeprecision) << setw(largewidth);
|
445 : |
|
|
|
446 : |
|
|
//JMB crude filter to remove the 'silly' values from the output
|
447 : |
|
|
if ((*modelDistribution[timeindex][area])[age][len] < rathersmall)
|
448 : |
|
|
outfile << 0 << endl;
|
449 : |
|
|
else
|
450 : |
|
|
outfile << (*modelDistribution[timeindex][area])[age][len] << endl;
|
451 : |
|
|
}
|
452 : |
|
|
}
|
453 : |
|
|
}
|
454 : |
|
|
}
|
455 : |
|
|
|
456 : |
|
|
void CatchDistribution::setFleetsAndStocks(FleetPtrVector& Fleets, StockPtrVector& Stocks) {
|
457 : |
|
|
int i, j, k, found, minage, maxage;
|
458 : |
|
|
FleetPtrVector fleets;
|
459 : |
|
|
StockPtrVector stocks;
|
460 : |
|
|
|
461 : |
|
|
for (i = 0; i < fleetnames.Size(); i++) {
|
462 : |
|
|
found = 0;
|
463 : |
|
|
for (j = 0; j < Fleets.Size(); j++) {
|
464 : |
|
|
if (strcasecmp(fleetnames[i], Fleets[j]->getName()) == 0) {
|
465 : |
|
|
found ++;
|
466 : |
|
|
fleets.resize(Fleets[j]);
|
467 : |
|
|
}
|
468 : |
|
|
}
|
469 : |
|
|
if (found == 0)
|
470 : |
|
|
handle.logMessage(LOGFAIL, "Error in catchdistribution - unrecognised fleet", fleetnames[i]);
|
471 : |
|
|
}
|
472 : |
|
|
|
473 : |
|
|
for (i = 0; i < fleets.Size(); i++)
|
474 : |
|
|
for (j = 0; j < fleets.Size(); j++)
|
475 : |
|
|
if ((strcasecmp(fleets[i]->getName(), fleets[j]->getName()) == 0) && (i != j))
|
476 : |
|
|
handle.logMessage(LOGFAIL, "Error in catchdistribution - repeated fleet", fleets[i]->getName());
|
477 : |
|
|
|
478 : |
|
|
for (i = 0; i < stocknames.Size(); i++) {
|
479 : |
|
|
found = 0;
|
480 : |
|
|
for (j = 0; j < Stocks.Size(); j++) {
|
481 : |
|
|
if (Stocks[j]->isEaten()) {
|
482 : |
|
|
if (strcasecmp(stocknames[i], Stocks[j]->getName()) == 0) {
|
483 : |
|
|
found++;
|
484 : |
|
|
stocks.resize(Stocks[j]);
|
485 : |
|
|
}
|
486 : |
|
|
}
|
487 : |
|
|
}
|
488 : |
|
|
if (found == 0)
|
489 : |
|
|
handle.logMessage(LOGFAIL, "Error in catchdistribution - unrecognised stock", stocknames[i]);
|
490 : |
|
|
}
|
491 : |
|
|
|
492 : |
|
|
for (i = 0; i < stocks.Size(); i++)
|
493 : |
|
|
for (j = 0; j < stocks.Size(); j++)
|
494 : |
|
|
if ((strcasecmp(stocks[i]->getName(), stocks[j]->getName()) == 0) && (i != j))
|
495 : |
|
|
handle.logMessage(LOGFAIL, "Error in catchdistribution - repeated stock", stocks[i]->getName());
|
496 : |
|
|
|
497 : |
|
|
//check fleet areas and stock areas, ages and lengths
|
498 : |
|
|
if (handle.getLogLevel() >= LOGWARN) {
|
499 : |
|
|
for (j = 0; j < areas.Nrow(); j++) {
|
500 : |
|
|
found = 0;
|
501 : |
|
|
for (i = 0; i < fleets.Size(); i++)
|
502 : |
|
|
for (k = 0; k < areas.Ncol(j); k++)
|
503 : |
|
|
if (fleets[i]->isInArea(areas[j][k]))
|
504 : |
|
|
found++;
|
505 : |
|
|
if (found == 0)
|
506 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - fleet not defined on all areas");
|
507 : |
|
|
}
|
508 : |
|
|
|
509 : |
|
|
for (j = 0; j < areas.Nrow(); j++) {
|
510 : |
|
|
found = 0;
|
511 : |
|
|
for (i = 0; i < stocks.Size(); i++)
|
512 : |
|
|
for (k = 0; k < areas.Ncol(j); k++)
|
513 : |
|
|
if (stocks[i]->isInArea(areas[j][k]))
|
514 : |
|
|
found++;
|
515 : |
|
|
if (found == 0)
|
516 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - stock not defined on all areas");
|
517 : |
|
|
}
|
518 : |
|
|
|
519 : |
|
|
minage = 9999;
|
520 : |
|
|
maxage = 0;
|
521 : |
|
|
for (i = 0; i < ages.Nrow(); i++) {
|
522 : |
|
|
for (j = 0; j < ages.Ncol(i); j++) {
|
523 : |
|
|
minage = min(ages[i][j], minage);
|
524 : |
|
|
maxage = max(ages[i][j], maxage);
|
525 : |
|
|
}
|
526 : |
|
|
}
|
527 : |
|
|
|
528 : |
|
|
found = 0;
|
529 : |
|
|
for (i = 0; i < stocks.Size(); i++)
|
530 : |
|
|
if (minage >= stocks[i]->minAge())
|
531 : |
|
|
found++;
|
532 : |
|
|
if (found == 0)
|
533 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - minimum age less than stock age");
|
534 : |
|
|
|
535 : |
|
|
found = 0;
|
536 : |
|
|
for (i = 0; i < stocks.Size(); i++)
|
537 : |
|
|
if (maxage <= stocks[i]->maxAge())
|
538 : |
|
|
found++;
|
539 : |
|
|
if (found == 0)
|
540 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - maximum age greater than stock age");
|
541 : |
|
|
|
542 : |
|
|
found = 0;
|
543 : |
|
|
for (i = 0; i < stocks.Size(); i++)
|
544 : |
|
|
if (LgrpDiv->maxLength(0) > stocks[i]->getLengthGroupDiv()->minLength())
|
545 : |
|
|
found++;
|
546 : |
|
|
if (found == 0)
|
547 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - minimum length group less than stock length");
|
548 : |
|
|
|
549 : |
|
|
found = 0;
|
550 : |
|
|
for (i = 0; i < stocks.Size(); i++)
|
551 : |
|
|
if (LgrpDiv->minLength(LgrpDiv->numLengthGroups()) < stocks[i]->getLengthGroupDiv()->maxLength())
|
552 : |
|
|
found++;
|
553 : |
|
|
if (found == 0)
|
554 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - maximum length group greater than stock length");
|
555 : |
|
|
}
|
556 : |
|
|
|
557 : |
|
|
aggregator = new FleetPreyAggregator(fleets, stocks, LgrpDiv, areas, ages, overconsumption);
|
558 : |
|
|
}
|
559 : |
|
|
|
560 : |
|
|
void CatchDistribution::addLikelihood(const TimeClass* const TimeInfo) {
|
561 : |
|
|
|
562 : |
|
|
if ((!(AAT.atCurrentTime(TimeInfo))) || (isZero(weight)))
|
563 : |
|
|
return;
|
564 : |
|
|
|
565 : |
|
|
if ((handle.getLogLevel() >= LOGMESSAGE) && ((!yearly) || (TimeInfo->getStep() == TimeInfo->numSteps())))
|
566 : |
|
|
handle.logMessage(LOGMESSAGE, "Calculating likelihood score for catchdistribution component", this->getName());
|
567 : |
|
|
|
568 : |
|
|
int i;
|
569 : |
|
|
timeindex = -1;
|
570 : |
|
|
for (i = 0; i < Years.Size(); i++)
|
571 : |
|
|
if ((Years[i] == TimeInfo->getYear()) && (Steps[i] == TimeInfo->getStep()))
|
572 : |
|
|
timeindex = i;
|
573 : |
|
|
if (timeindex == -1)
|
574 : |
|
|
handle.logMessage(LOGFAIL, "Error in catchdistribution - invalid timestep");
|
575 : |
|
|
|
576 : |
|
|
aggregator->Sum();
|
577 : |
|
|
if ((handle.getLogLevel() >= LOGWARN) && (aggregator->checkCatchData()))
|
578 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - zero catch found");
|
579 : |
|
|
alptr = &aggregator->getSum();
|
580 : |
|
|
|
581 : |
|
|
double l = 0.0;
|
582 : |
|
|
switch (functionnumber) {
|
583 : |
|
|
case 1:
|
584 : |
|
|
l = calcLikMultinomial();
|
585 : |
|
|
break;
|
586 : |
|
|
case 2:
|
587 : |
|
|
l = calcLikPearson(TimeInfo);
|
588 : |
|
|
break;
|
589 : |
|
|
case 3:
|
590 : |
|
|
l = calcLikGamma(TimeInfo);
|
591 : |
|
|
break;
|
592 : |
|
|
case 4:
|
593 : |
|
|
l = calcLikSumSquares(TimeInfo);
|
594 : |
|
|
break;
|
595 : |
|
|
case 5:
|
596 : |
|
|
l = calcLikMVNormal();
|
597 : |
|
|
break;
|
598 : |
|
|
case 6:
|
599 : |
|
|
l = calcLikMVLogistic();
|
600 : |
|
|
break;
|
601 : |
|
|
case 7:
|
602 : |
|
|
l = calcLikLog(TimeInfo);
|
603 : |
|
|
break;
|
604 : |
|
|
case 8:
|
605 : |
|
|
l = calcLikStratified(TimeInfo);
|
606 : |
|
|
break;
|
607 : |
|
|
default:
|
608 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - unrecognised function", functionname);
|
609 : |
|
|
break;
|
610 : |
|
|
}
|
611 : |
|
|
|
612 : |
|
|
if ((!yearly) || (TimeInfo->getStep() == TimeInfo->numSteps())) {
|
613 : |
|
|
likelihood += l;
|
614 : |
|
|
if (handle.getLogLevel() >= LOGMESSAGE)
|
615 : |
|
|
handle.logMessage(LOGMESSAGE, "The likelihood score for this component on this timestep is", l);
|
616 : |
|
|
}
|
617 : |
|
|
}
|
618 : |
|
|
|
619 : |
|
|
double CatchDistribution::calcLikMultinomial() {
|
620 : |
|
|
int area, age, len;
|
621 : |
|
|
int numage = ages.Nrow();
|
622 : |
|
|
int numlen = LgrpDiv->numLengthGroups();
|
623 : |
|
|
DoubleVector dist(numage, 0.0);
|
624 : |
|
|
DoubleVector data(numage, 0.0);
|
625 : |
|
|
|
626 : |
|
|
MN.Reset();
|
627 : |
|
|
//the object MN does most of the work, accumulating likelihood
|
628 : |
|
|
for (area = 0; area < areas.Nrow(); area++) {
|
629 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
630 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++)
|
631 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++)
|
632 : |
|
|
(*modelDistribution[timeindex][area])[age][len] = ((*alptr)[area][age][len]).N;
|
633 : |
|
|
|
634 : |
|
|
if (numage == 1) {
|
635 : |
|
|
//only one age-group, so calculate multinomial based on length distribution
|
636 : |
|
|
likelihoodValues[timeindex][area] +=
|
637 : |
|
|
MN.calcLogLikelihood((*obsDistribution[timeindex][area])[0],
|
638 : |
|
|
(*modelDistribution[timeindex][area])[0]);
|
639 : |
|
|
|
640 : |
|
|
} else {
|
641 : |
|
|
//many age-groups, so calculate multinomial based on age distribution per length group
|
642 : |
|
|
for (len = 0; len < numlen; len++) {
|
643 : |
|
|
for (age = 0; age < numage; age++) {
|
644 : |
|
|
dist[age] = (*modelDistribution[timeindex][area])[age][len];
|
645 : |
|
|
data[age] = (*obsDistribution[timeindex][area])[age][len];
|
646 : |
|
|
}
|
647 : |
|
|
likelihoodValues[timeindex][area] += MN.calcLogLikelihood(data, dist);
|
648 : |
|
|
}
|
649 : |
|
|
}
|
650 : |
|
|
}
|
651 : |
|
|
return MN.getLogLikelihood();
|
652 : |
|
|
}
|
653 : |
|
|
|
654 : |
|
|
double CatchDistribution::calcLikPearson(const TimeClass* const TimeInfo) {
|
655 : |
|
|
/* written by Hoskuldur Bjornsson 29/8 98
|
656 : |
|
|
* corrected by kgf 16/9 98
|
657 : |
|
|
* modified by kgf 11/11 98 to make it possible to sum up catches
|
658 : |
|
|
* and calculated catches on year basis.
|
659 : |
|
|
* Modified 3/5 99 by kgf to check the age intervals of the stock
|
660 : |
|
|
* and the catch data, and make use of the ages that are common
|
661 : |
|
|
* for the stock and the catch data.*/
|
662 : |
|
|
|
663 : |
|
|
double totallikelihood = 0.0;
|
664 : |
|
|
int age, len, area;
|
665 : |
|
|
|
666 : |
|
|
for (area = 0; area < areas.Nrow(); area++) {
|
667 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
668 : |
|
|
|
669 : |
|
|
//JMB - changed to remove the need to store minrow and mincol stuff ...
|
670 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++)
|
671 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++)
|
672 : |
|
|
(*modelDistribution[timeindex][area])[age][len] = (*alptr)[area][age][len].N;
|
673 : |
|
|
|
674 : |
|
|
if (!yearly) { //calculate likelihood on all steps
|
675 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
676 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
677 : |
|
|
likelihoodValues[timeindex][area] +=
|
678 : |
|
|
((*modelDistribution[timeindex][area])[age][len] -
|
679 : |
|
|
(*obsDistribution[timeindex][area])[age][len]) *
|
680 : |
|
|
((*modelDistribution[timeindex][area])[age][len] -
|
681 : |
|
|
(*obsDistribution[timeindex][area])[age][len]) /
|
682 : |
|
|
fabs(((*modelDistribution[timeindex][area])[age][len] + epsilon));
|
683 : |
|
|
}
|
684 : |
|
|
}
|
685 : |
|
|
totallikelihood += likelihoodValues[timeindex][area];
|
686 : |
|
|
|
687 : |
|
|
} else { //calculate likelihood on year basis
|
688 : |
|
|
|
689 : |
|
|
if (TimeInfo->getStep() == 1) { //start of a new year
|
690 : |
|
|
(*modelYearData[area]).setToZero();
|
691 : |
|
|
(*obsYearData[area]).setToZero();
|
692 : |
|
|
}
|
693 : |
|
|
|
694 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
695 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
696 : |
|
|
(*modelYearData[area])[age][len] += (*modelDistribution[timeindex][area])[age][len];
|
697 : |
|
|
(*obsYearData[area])[age][len] += (*obsDistribution[timeindex][area])[age][len];
|
698 : |
|
|
}
|
699 : |
|
|
}
|
700 : |
|
|
|
701 : |
|
|
if (TimeInfo->getStep() < TimeInfo->numSteps())
|
702 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
703 : |
|
|
else { //last step in year, so need to calc likelihood contribution
|
704 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
705 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
706 : |
|
|
likelihoodValues[timeindex][area] +=
|
707 : |
|
|
((*modelYearData[area])[age][len] - (*obsYearData[area])[age][len]) *
|
708 : |
|
|
((*modelYearData[area])[age][len] - (*obsYearData[area])[age][len]) /
|
709 : |
|
|
fabs(((*modelYearData[area])[age][len] + epsilon));
|
710 : |
|
|
}
|
711 : |
|
|
}
|
712 : |
|
|
totallikelihood += likelihoodValues[timeindex][area];
|
713 : |
|
|
}
|
714 : |
|
|
}
|
715 : |
|
|
}
|
716 : |
|
|
return totallikelihood;
|
717 : |
|
|
}
|
718 : |
|
|
|
719 : |
|
|
double CatchDistribution::calcLikGamma(const TimeClass* const TimeInfo) {
|
720 : |
|
|
//written kgf 24/5 00
|
721 : |
|
|
//Formula by Hans J Skaug, 15/3 00 No weighting at present.
|
722 : |
|
|
//This function is scale independent.
|
723 : |
|
|
|
724 : |
|
|
double totallikelihood = 0.0;
|
725 : |
|
|
int age, len, area;
|
726 : |
|
|
|
727 : |
|
|
for (area = 0; area < areas.Nrow(); area++) {
|
728 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
729 : |
|
|
|
730 : |
|
|
//JMB - changed to remove the need to store minrow and mincol stuff ...
|
731 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++)
|
732 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++)
|
733 : |
|
|
(*modelDistribution[timeindex][area])[age][len] = (*alptr)[area][age][len].N;
|
734 : |
|
|
|
735 : |
|
|
if (!yearly) { //calculate likelihood on all steps
|
736 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
737 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
738 : |
|
|
likelihoodValues[timeindex][area] +=
|
739 : |
|
|
(*obsDistribution[timeindex][area])[age][len] /
|
740 : |
|
|
((*modelDistribution[timeindex][area])[age][len] + epsilon) +
|
741 : |
|
|
log((*modelDistribution[timeindex][area])[age][len] + epsilon);
|
742 : |
|
|
}
|
743 : |
|
|
}
|
744 : |
|
|
totallikelihood += likelihoodValues[timeindex][area];
|
745 : |
|
|
|
746 : |
|
|
} else { //calculate likelihood on year basis
|
747 : |
|
|
|
748 : |
|
|
if (TimeInfo->getStep() == 1) { //start of a new year
|
749 : |
|
|
(*modelYearData[area]).setToZero();
|
750 : |
|
|
(*obsYearData[area]).setToZero();
|
751 : |
|
|
}
|
752 : |
|
|
|
753 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
754 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
755 : |
|
|
(*modelYearData[area])[age][len] += (*modelDistribution[timeindex][area])[age][len];
|
756 : |
|
|
(*obsYearData[area])[age][len] += (*obsDistribution[timeindex][area])[age][len];
|
757 : |
|
|
}
|
758 : |
|
|
}
|
759 : |
|
|
|
760 : |
|
|
if (TimeInfo->getStep() < TimeInfo->numSteps())
|
761 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
762 : |
|
|
else { //last step in year, so need to calc likelihood contribution
|
763 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
764 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
765 : |
|
|
likelihoodValues[timeindex][area] +=
|
766 : |
|
|
(*obsYearData[area])[age][len] / ((*modelYearData[area])[age][len] + epsilon) +
|
767 : |
|
|
log((*modelYearData[area])[age][len] + epsilon);
|
768 : |
|
|
}
|
769 : |
|
|
}
|
770 : |
|
|
totallikelihood += likelihoodValues[timeindex][area];
|
771 : |
|
|
}
|
772 : |
|
|
}
|
773 : |
|
|
}
|
774 : |
|
|
return totallikelihood;
|
775 : |
|
|
}
|
776 : |
|
|
|
777 : |
|
|
double CatchDistribution::calcLikLog(const TimeClass* const TimeInfo) {
|
778 : |
|
|
//written by kgf 23/11 98 to get a better scaling of the stocks.
|
779 : |
|
|
//modified by kgf 27/11 98 to sum first and then take the logarithm
|
780 : |
|
|
|
781 : |
|
|
double totallikelihood = 0.0;
|
782 : |
|
|
int area, age, len;
|
783 : |
|
|
double totalmodel, totaldata, ratio;
|
784 : |
|
|
|
785 : |
|
|
for (area = 0; area < areas.Nrow(); area++) {
|
786 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
787 : |
|
|
totalmodel = 0.0;
|
788 : |
|
|
totaldata = 0.0;
|
789 : |
|
|
|
790 : |
|
|
//JMB - changed to remove the need to store minrow and mincol stuff ...
|
791 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++)
|
792 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++)
|
793 : |
|
|
(*modelDistribution[timeindex][area])[age][len] = (*alptr)[area][age][len].N;
|
794 : |
|
|
|
795 : |
|
|
if (!yearly) { //calculate likelihood on all steps
|
796 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
797 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
798 : |
|
|
totalmodel += (*modelDistribution[timeindex][area])[age][len];
|
799 : |
|
|
totaldata += (*obsDistribution[timeindex][area])[age][len];
|
800 : |
|
|
}
|
801 : |
|
|
}
|
802 : |
|
|
ratio = log(totaldata / totalmodel);
|
803 : |
|
|
likelihoodValues[timeindex][area] += (ratio * ratio);
|
804 : |
|
|
|
805 : |
|
|
} else { //calculate likelihood on year basis
|
806 : |
|
|
|
807 : |
|
|
if (TimeInfo->getStep() == 1) { //start of a new year
|
808 : |
|
|
(*modelYearData[area]).setToZero();
|
809 : |
|
|
(*obsYearData[area]).setToZero();
|
810 : |
|
|
}
|
811 : |
|
|
|
812 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
813 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
814 : |
|
|
(*modelYearData[area])[age][len] += (*modelDistribution[timeindex][area])[age][len];
|
815 : |
|
|
(*obsYearData[area])[age][len] += (*obsDistribution[timeindex][area])[age][len];
|
816 : |
|
|
}
|
817 : |
|
|
}
|
818 : |
|
|
|
819 : |
|
|
if (TimeInfo->getStep() < TimeInfo->numSteps())
|
820 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
821 : |
|
|
else { //last step in year, so need to calculate likelihood contribution
|
822 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
823 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
824 : |
|
|
totalmodel += (*modelYearData[area])[age][len];
|
825 : |
|
|
totaldata += (*obsYearData[area])[age][len];
|
826 : |
|
|
}
|
827 : |
|
|
}
|
828 : |
|
|
ratio = log(totaldata / totalmodel);
|
829 : |
|
|
likelihoodValues[timeindex][area] += (ratio * ratio);
|
830 : |
|
|
}
|
831 : |
|
|
}
|
832 : |
|
|
totallikelihood += likelihoodValues[timeindex][area];
|
833 : |
|
|
}
|
834 : |
|
|
return totallikelihood;
|
835 : |
|
|
}
|
836 : |
|
|
|
837 : |
|
|
double CatchDistribution::calcLikSumSquares(const TimeClass* const TimeInfo) {
|
838 : |
|
|
|
839 : |
|
|
double temp, totallikelihood, totalmodel, totaldata;
|
840 : |
|
|
int age, len, area;
|
841 : |
|
|
|
842 : |
|
|
totallikelihood = 0.0;
|
843 : |
|
|
for (area = 0; area < areas.Nrow(); area++) {
|
844 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
845 : |
|
|
|
846 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++)
|
847 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++)
|
848 : |
|
|
(*modelDistribution[timeindex][area])[age][len] = ((*alptr)[area][age][len]).N;
|
849 : |
|
|
|
850 : |
|
|
totalmodel = 0.0;
|
851 : |
|
|
totaldata = 0.0;
|
852 : |
|
|
if (!yearly) { //calculate likelihood on all steps
|
853 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
854 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
855 : |
|
|
totalmodel += (*modelDistribution[timeindex][area])[age][len];
|
856 : |
|
|
totaldata += (*obsDistribution[timeindex][area])[age][len];
|
857 : |
|
|
}
|
858 : |
|
|
}
|
859 : |
|
|
|
860 : |
|
|
if (!(isZero(totalmodel)))
|
861 : |
|
|
totalmodel = 1.0 / totalmodel;
|
862 : |
|
|
if (!(isZero(totaldata)))
|
863 : |
|
|
totaldata = 1.0 / totaldata;
|
864 : |
|
|
|
865 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
866 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
867 : |
|
|
temp = (((*obsDistribution[timeindex][area])[age][len] * totaldata)
|
868 : |
|
|
- ((*modelDistribution[timeindex][area])[age][len] * totalmodel));
|
869 : |
|
|
likelihoodValues[timeindex][area] += (temp * temp);
|
870 : |
|
|
}
|
871 : |
|
|
}
|
872 : |
|
|
totallikelihood += likelihoodValues[timeindex][area];
|
873 : |
|
|
|
874 : |
|
|
} else { //calculate likelihood on year basis
|
875 : |
|
|
|
876 : |
|
|
if (TimeInfo->getStep() == 1) { //start of a new year
|
877 : |
|
|
(*modelYearData[area]).setToZero();
|
878 : |
|
|
(*obsYearData[area]).setToZero();
|
879 : |
|
|
}
|
880 : |
|
|
|
881 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
882 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
883 : |
|
|
(*modelYearData[area])[age][len] += (*modelDistribution[timeindex][area])[age][len];
|
884 : |
|
|
(*obsYearData[area])[age][len] += (*obsDistribution[timeindex][area])[age][len];
|
885 : |
|
|
}
|
886 : |
|
|
}
|
887 : |
|
|
|
888 : |
|
|
if (TimeInfo->getStep() < TimeInfo->numSteps())
|
889 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
890 : |
|
|
else { //last step in year, so need to calculate likelihood contribution
|
891 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
892 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
893 : |
|
|
totalmodel += (*modelYearData[area])[age][len];
|
894 : |
|
|
totaldata += (*obsYearData[area])[age][len];
|
895 : |
|
|
}
|
896 : |
|
|
}
|
897 : |
|
|
|
898 : |
|
|
if (!(isZero(totalmodel)))
|
899 : |
|
|
totalmodel = 1.0 / totalmodel;
|
900 : |
|
|
if (!(isZero(totaldata)))
|
901 : |
|
|
totaldata = 1.0 / totaldata;
|
902 : |
|
|
|
903 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
904 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
905 : |
|
|
temp = (((*obsYearData[area])[age][len] * totaldata)
|
906 : |
|
|
- ((*modelYearData[area])[age][len] * totalmodel));
|
907 : |
|
|
likelihoodValues[timeindex][area] += (temp * temp);
|
908 : |
|
|
}
|
909 : |
|
|
}
|
910 : |
|
|
totallikelihood += likelihoodValues[timeindex][area];
|
911 : |
|
|
}
|
912 : |
|
|
}
|
913 : |
|
|
}
|
914 : |
|
|
return totallikelihood;
|
915 : |
|
|
}
|
916 : |
|
|
|
917 : |
|
|
double CatchDistribution::calcLikStratified(const TimeClass* const TimeInfo) {
|
918 : |
|
|
|
919 : |
|
|
int numage = ages.Nrow();
|
920 : |
|
|
int numlen = LgrpDiv->numLengthGroups();
|
921 : |
|
|
double temp, totallikelihood, totalmodel, totaldata;
|
922 : |
|
|
int age, len, area;
|
923 : |
|
|
|
924 : |
|
|
totallikelihood = 0.0;
|
925 : |
|
|
for (area = 0; area < areas.Nrow(); area++) {
|
926 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
927 : |
|
|
|
928 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++)
|
929 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++)
|
930 : |
|
|
(*modelDistribution[timeindex][area])[age][len] = ((*alptr)[area][age][len]).N;
|
931 : |
|
|
|
932 : |
|
|
if (!yearly) { //calculate likelihood on all steps
|
933 : |
|
|
//calculate an age distribution for each length class
|
934 : |
|
|
for (len = 0; len < numlen; len++) {
|
935 : |
|
|
totalmodel = 0.0;
|
936 : |
|
|
totaldata = 0.0;
|
937 : |
|
|
for (age = 0; age < numage; age++) {
|
938 : |
|
|
totalmodel += (*modelDistribution[timeindex][area])[age][len];
|
939 : |
|
|
totaldata += (*obsDistribution[timeindex][area])[age][len];
|
940 : |
|
|
}
|
941 : |
|
|
|
942 : |
|
|
if (!(isZero(totalmodel)))
|
943 : |
|
|
totalmodel = 1.0 / totalmodel;
|
944 : |
|
|
if (!(isZero(totaldata)))
|
945 : |
|
|
totaldata = 1.0 / totaldata;
|
946 : |
|
|
|
947 : |
|
|
for (age = 0; age < numage; age++) {
|
948 : |
|
|
temp = (((*obsDistribution[timeindex][area])[age][len] * totaldata)
|
949 : |
|
|
- ((*modelDistribution[timeindex][area])[age][len] * totalmodel));
|
950 : |
|
|
likelihoodValues[timeindex][area] += (temp * temp);
|
951 : |
|
|
}
|
952 : |
|
|
}
|
953 : |
|
|
totallikelihood += likelihoodValues[timeindex][area];
|
954 : |
|
|
|
955 : |
|
|
} else { //calculate likelihood on year basis
|
956 : |
|
|
|
957 : |
|
|
if (TimeInfo->getStep() == 1) { //start of a new year
|
958 : |
|
|
(*modelYearData[area]).setToZero();
|
959 : |
|
|
(*obsYearData[area]).setToZero();
|
960 : |
|
|
}
|
961 : |
|
|
|
962 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
963 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
964 : |
|
|
(*modelYearData[area])[age][len] += (*modelDistribution[timeindex][area])[age][len];
|
965 : |
|
|
(*obsYearData[area])[age][len] += (*obsDistribution[timeindex][area])[age][len];
|
966 : |
|
|
}
|
967 : |
|
|
}
|
968 : |
|
|
|
969 : |
|
|
if (TimeInfo->getStep() < TimeInfo->numSteps())
|
970 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
971 : |
|
|
else { //last step in year, so need to calculate likelihood contribution
|
972 : |
|
|
//calculate an age distribution for each length class
|
973 : |
|
|
for (len = 0; len < numlen; len++) {
|
974 : |
|
|
totalmodel = 0.0;
|
975 : |
|
|
totaldata = 0.0;
|
976 : |
|
|
for (age = 0; age < numage; age++) {
|
977 : |
|
|
totalmodel += (*modelYearData[area])[age][len];
|
978 : |
|
|
totaldata += (*obsYearData[area])[age][len];
|
979 : |
|
|
}
|
980 : |
|
|
|
981 : |
|
|
if (!(isZero(totalmodel)))
|
982 : |
|
|
totalmodel = 1.0 / totalmodel;
|
983 : |
|
|
if (!(isZero(totaldata)))
|
984 : |
|
|
totaldata = 1.0 / totaldata;
|
985 : |
|
|
|
986 : |
|
|
for (age = 0; age < numage; age++) {
|
987 : |
|
|
temp = (((*obsYearData[area])[age][len] * totaldata)
|
988 : |
|
|
- ((*modelYearData[area])[age][len] * totalmodel));
|
989 : |
|
|
likelihoodValues[timeindex][area] += (temp * temp);
|
990 : |
|
|
}
|
991 : |
|
|
}
|
992 : |
|
|
totallikelihood += likelihoodValues[timeindex][area];
|
993 : |
|
|
}
|
994 : |
|
|
}
|
995 : |
|
|
}
|
996 : |
|
|
return totallikelihood;
|
997 : |
|
|
}
|
998 : |
|
|
|
999 : |
|
|
void CatchDistribution::calcCorrelation() {
|
1000 : |
|
|
int i, j, l, p;
|
1001 : |
|
|
p = LgrpDiv->numLengthGroups();
|
1002 : |
|
|
DoubleMatrix correlation(p, p, 0.0);
|
1003 : |
|
|
|
1004 : |
|
|
for (i = 0; i < lag; i++)
|
1005 : |
|
|
if (fabs(params[i] - 1.0) > 1.0)
|
1006 : |
|
|
illegal = 1;
|
1007 : |
|
|
|
1008 : |
|
|
if (!illegal) {
|
1009 : |
|
|
for (i = 0; i < p; i++) {
|
1010 : |
|
|
for (j = 0; j <= i; j++) {
|
1011 : |
|
|
for (l = 1; l <= lag; l++) {
|
1012 : |
|
|
if ((i - l) >= 0) {
|
1013 : |
|
|
correlation[i][j] += (params[l - 1] - 1.0) * correlation[i - l][j];
|
1014 : |
|
|
correlation[j][i] += (params[l - 1] - 1.0) * correlation[i - l][j];
|
1015 : |
|
|
}
|
1016 : |
|
|
}
|
1017 : |
|
|
}
|
1018 : |
|
|
correlation[i][i] += sigma * sigma;
|
1019 : |
|
|
}
|
1020 : |
|
|
LU = LUDecomposition(correlation);
|
1021 : |
|
|
}
|
1022 : |
|
|
}
|
1023 : |
|
|
|
1024 : |
|
|
|
1025 : |
|
|
double CatchDistribution::calcLikMVNormal() {
|
1026 : |
|
|
|
1027 : |
|
|
double totallikelihood = 0.0;
|
1028 : |
|
|
double sumdata, sumdist;
|
1029 : |
|
|
int age, len, area;
|
1030 : |
|
|
|
1031 : |
|
|
if ((illegal) || (LU.isIllegal()) || isZero(sigma))
|
1032 : |
|
|
return verybig;
|
1033 : |
|
|
|
1034 : |
|
|
DoubleVector diff(LgrpDiv->numLengthGroups(), 0.0);
|
1035 : |
|
|
for (area = 0; area < areas.Nrow(); area++) {
|
1036 : |
|
|
sumdata = 0.0;
|
1037 : |
|
|
sumdist = 0.0;
|
1038 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
1039 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
1040 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
1041 : |
|
|
(*modelDistribution[timeindex][area])[age][len] = ((*alptr)[area][age][len]).N;
|
1042 : |
|
|
sumdata += (*obsDistribution[timeindex][area])[age][len];
|
1043 : |
|
|
sumdist += (*modelDistribution[timeindex][area])[age][len];
|
1044 : |
|
|
}
|
1045 : |
|
|
}
|
1046 : |
|
|
|
1047 : |
|
|
if (isZero(sumdata))
|
1048 : |
|
|
sumdata = verybig;
|
1049 : |
|
|
else
|
1050 : |
|
|
sumdata = 1.0 / sumdata;
|
1051 : |
|
|
if (isZero(sumdist))
|
1052 : |
|
|
sumdist = verybig;
|
1053 : |
|
|
else
|
1054 : |
|
|
sumdist = 1.0 / sumdist;
|
1055 : |
|
|
|
1056 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
1057 : |
|
|
diff.setToZero();
|
1058 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++)
|
1059 : |
|
|
diff[len] = ((*obsDistribution[timeindex][area])[age][len] * sumdata)
|
1060 : |
|
|
- ((*modelDistribution[timeindex][area])[age][len] * sumdist);
|
1061 : |
|
|
|
1062 : |
|
|
likelihoodValues[timeindex][area] += diff * LU.Solve(diff);
|
1063 : |
|
|
}
|
1064 : |
|
|
totallikelihood += likelihoodValues[timeindex][area];
|
1065 : |
|
|
}
|
1066 : |
|
|
|
1067 : |
|
|
totallikelihood += LU.getLogDet() * alptr->Size();
|
1068 : |
|
|
return totallikelihood;
|
1069 : |
|
|
}
|
1070 : |
|
|
|
1071 : |
|
|
double CatchDistribution::calcLikMVLogistic() {
|
1072 : |
|
|
|
1073 : |
|
|
double totallikelihood = 0.0;
|
1074 : |
|
|
double sumdata = 0.0, sumdist = 0.0, sumnu = 0.0;
|
1075 : |
|
|
int age, len, area, p;
|
1076 : |
|
|
|
1077 : |
|
|
p = LgrpDiv->numLengthGroups();
|
1078 : |
|
|
DoubleVector nu(p, 0.0);
|
1079 : |
|
|
|
1080 : |
|
|
for (area = 0; area < areas.Nrow(); area++) {
|
1081 : |
|
|
likelihoodValues[timeindex][area] = 0.0;
|
1082 : |
|
|
for (age = (*alptr)[area].minAge(); age <= (*alptr)[area].maxAge(); age++) {
|
1083 : |
|
|
sumdata = 0.0;
|
1084 : |
|
|
sumdist = 0.0;
|
1085 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
1086 : |
|
|
(*modelDistribution[timeindex][area])[age][len] = ((*alptr)[area][age][len]).N;
|
1087 : |
|
|
sumdata += (*obsDistribution[timeindex][area])[age][len];
|
1088 : |
|
|
sumdist += (*modelDistribution[timeindex][area])[age][len];
|
1089 : |
|
|
}
|
1090 : |
|
|
|
1091 : |
|
|
if (isZero(sumdata))
|
1092 : |
|
|
sumdata = verybig;
|
1093 : |
|
|
else
|
1094 : |
|
|
sumdata = 1.0 / sumdata;
|
1095 : |
|
|
if (isZero(sumdist))
|
1096 : |
|
|
sumdist = verybig;
|
1097 : |
|
|
else
|
1098 : |
|
|
sumdist = 1.0 / sumdist;
|
1099 : |
|
|
|
1100 : |
|
|
sumnu = 0.0;
|
1101 : |
|
|
nu.setToZero();
|
1102 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++) {
|
1103 : |
|
|
nu[len] = log(((*obsDistribution[timeindex][area])[age][len] * sumdata) + verysmall)
|
1104 : |
|
|
- log(((*modelDistribution[timeindex][area])[age][len] * sumdist) + verysmall);
|
1105 : |
|
|
|
1106 : |
|
|
sumnu += nu[len];
|
1107 : |
|
|
}
|
1108 : |
|
|
sumnu = sumnu / p;
|
1109 : |
|
|
|
1110 : |
|
|
for (len = (*alptr)[area].minLength(age); len < (*alptr)[area].maxLength(age); len++)
|
1111 : |
|
|
likelihoodValues[timeindex][area] += (nu[len] - sumnu) * (nu[len] - sumnu);
|
1112 : |
|
|
}
|
1113 : |
|
|
totallikelihood += likelihoodValues[timeindex][area];
|
1114 : |
|
|
}
|
1115 : |
|
|
|
1116 : |
|
|
if (isZero(sigma)) {
|
1117 : |
|
|
handle.logMessage(LOGWARN, "Warning in catchdistribution - multivariate logistic sigma is zero");
|
1118 : |
|
|
return verybig;
|
1119 : |
|
|
}
|
1120 : |
|
|
|
1121 : |
|
|
totallikelihood = (totallikelihood / (sigma * sigma)) + (log(sigma) * (p - 1));
|
1122 : |
|
|
return totallikelihood;
|
1123 : |
|
|
}
|
1124 : |
|
|
|
1125 : |
|
|
void CatchDistribution::printSummary(ofstream& outfile) {
|
1126 : |
|
|
int year, area;
|
1127 : |
|
|
|
1128 : |
|
|
for (year = 0; year < likelihoodValues.Nrow(); year++) {
|
1129 : |
|
|
for (area = 0; area < likelihoodValues.Ncol(year); area++) {
|
1130 : |
|
|
if (!yearly) {
|
1131 : |
|
|
outfile << setw(lowwidth) << Years[year] << sep << setw(lowwidth)
|
1132 : |
|
|
<< Steps[year] << sep << setw(printwidth) << areaindex[area] << sep
|
1133 : |
|
|
<< setw(largewidth) << this->getName() << sep << setw(smallwidth) << weight
|
1134 : |
|
|
<< sep << setprecision(largeprecision) << setw(largewidth)
|
1135 : |
|
|
<< likelihoodValues[year][area] << endl;
|
1136 : |
|
|
} else {
|
1137 : |
|
|
if (isZero(likelihoodValues[year][area])) {
|
1138 : |
|
|
// assume that this isnt the last step for that year and ignore
|
1139 : |
|
|
} else {
|
1140 : |
|
|
outfile << setw(lowwidth) << Years[year] << " all "
|
1141 : |
|
|
<< setw(printwidth) << areaindex[area] << sep
|
1142 : |
|
|
<< setw(largewidth) << this->getName() << sep << setprecision(smallprecision)
|
1143 : |
|
|
<< setw(smallwidth) << weight << sep << setprecision(largeprecision)
|
1144 : |
|
|
<< setw(largewidth) << likelihoodValues[year][area] << endl;
|
1145 : |
|
|
}
|
1146 : |
|
|
}
|
1147 : |
|
|
}
|
1148 : |
|
|
}
|
1149 : |
|
|
outfile.flush();
|
1150 : |
|
|
}
|