Line data Source code
1 : //# StatWT.cc: Subtract the continuum from VisBuffGroups and
2 : //# write them to a different MS.
3 : //# Copyright (C) 2011
4 : //# Associated Universities, Inc. Washington DC, USA.
5 : //#
6 : //# This library is free software; you can redistribute it and/or modify it
7 : //# under the terms of the GNU Library General Public License as published by
8 : //# the Free Software Foundation; either version 2 of the License, or (at your
9 : //# option) any later version.
10 : //#
11 : //# This library is distributed in the hope that it will be useful, but WITHOUT
12 : //# ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 : //# FITNESS FOR A PARTICULAR PURPOSE. See the GNU Library General Public
14 : //# License for more details.
15 : //#
16 : //# You should have received a copy of the GNU Library General Public License
17 : //# along with this library; if not, write to the Free Software Foundation,
18 : //# Inc., 675 Massachusetts Ave, Cambridge, MA 02139, USA.
19 : //#
20 : //# Correspondence concerning AIPS++ should be addressed as follows:
21 : //# Internet email: casa-feedback@nrao.edu.
22 : //# Postal address: AIPS++ Project Office
23 : //# National Radio Astronomy Observatory
24 : //# 520 Edgemont Road
25 : //# Charlottesville, VA 22903-2475 USA
26 : //#
27 :
28 : #include <msvis/MSVis/StatWT.h>
29 : //#include <msvis/MSVis/SubMS.h>
30 : #include <msvis/MSVis/VisBufferComponents.h>
31 : #include <msvis/MSVis/VisBuffGroup.h>
32 : #include <msvis/MSVis/VisBuffGroupAcc.h>
33 : #include <casacore/casa/Exceptions/Error.h>
34 : #include <casacore/casa/Logging/LogIO.h>
35 : #include <casacore/ms/MSSel/MSSelection.h>
36 : #include <casacore/casa/Arrays/ArrayMath.h>
37 :
38 : using namespace casacore;
39 : namespace casa {
40 :
41 0 : StatWT::StatWT(const ROVisibilityIterator& vi,
42 : const MS::PredefinedColumns datacol,
43 : const String& fitspw,
44 : const String& outspw,
45 : const Bool dorms,
46 : const uInt minsamp,
47 0 : const vector<uInt> selcorrs) :
48 : GroupWorker(vi),
49 0 : datacol_p(datacol),
50 0 : fitspw_p(fitspw),
51 0 : outspw_p(outspw),
52 0 : dorms_p(dorms),
53 0 : rowsdone_p(0)
54 : {
55 0 : LogIO os(LogOrigin("StatWT", "StatWT()"));
56 :
57 0 : if(dorms && minsamp < 1){
58 : os << LogIO::WARN
59 : << "It takes at least one to measure an rms - using minsamp = 1."
60 0 : << LogIO::POST;
61 0 : minsamp_p = 1;
62 : }
63 0 : else if(!dorms && minsamp < 2){
64 : os << LogIO::WARN
65 : << "It takes at least two to measure a variance - using minsamp = 2."
66 0 : << LogIO::POST;
67 0 : minsamp_p = 2;
68 : }
69 : else
70 0 : minsamp_p = minsamp;
71 :
72 0 : prefetchColumns_p = asyncio::PrefetchColumns::prefetchColumns(
73 : VisBufferComponents::Ant1,
74 : VisBufferComponents::Ant2,
75 : VisBufferComponents::ArrayId,
76 : VisBufferComponents::CorrType,
77 : VisBufferComponents::DataDescriptionId,
78 : //VisBufferComponents::Feed1,
79 : //VisBufferComponents::Feed2,
80 : VisBufferComponents::FieldId,
81 : VisBufferComponents::FlagCube,
82 : VisBufferComponents::Flag,
83 : VisBufferComponents::FlagRow,
84 : VisBufferComponents::ObservationId,
85 : //VisBufferComponents::NChannel,
86 : VisBufferComponents::NCorr,
87 : VisBufferComponents::NRow,
88 : //VisBufferComponents::ProcessorId,
89 : VisBufferComponents::Scan,
90 : VisBufferComponents::SpW,
91 : VisBufferComponents::SigmaMat,
92 : VisBufferComponents::StateId,
93 : //VisBufferComponents::Time,
94 : //VisBufferComponents::TimeCentroid,
95 : //VisBufferComponents::TimeInterval,
96 : VisBufferComponents::WeightMat,
97 0 : -1);
98 0 : if(datacol == MS::DATA)
99 0 : prefetchColumns_p.insert(VisBufferComponents::ObservedCube);
100 0 : else if(datacol == MS::MODEL_DATA)
101 0 : prefetchColumns_p.insert(VisBufferComponents::ModelCube);
102 0 : else if(datacol == MS::CORRECTED_DATA)
103 0 : prefetchColumns_p.insert(VisBufferComponents::CorrectedCube);
104 : // else if(datacol == MS::FLOAT_DATA) // Not in VisBufferComponents yet.
105 : // prefetchColumns_p.insert(VisBufferComponents::FloatCube);
106 :
107 0 : VisBuffGroupAcc::fillChanMask(fitmask_p, fitspw, invi_p.ms());
108 :
109 0 : MSSelection mssel;
110 0 : mssel.setSpwExpr(outspw);
111 0 : Matrix<Int> chansel = mssel.getChanList(&(invi_p.ms()), 1);
112 0 : Vector<Int> spws(chansel.column(0));
113 0 : uInt nselspws = spws.nelements();
114 0 : selcorrs_p = selcorrs;
115 :
116 0 : for(uInt i = 0; i < nselspws; ++i)
117 0 : outspws_p.insert(spws[i]);
118 0 : }
119 :
120 0 : StatWT::~StatWT()
121 : {
122 0 : VisBuffGroupAcc::clearChanMask(fitmask_p);
123 0 : }
124 :
125 : // StatWT& StatWT::operator=(const StatWT &other)
126 : // {
127 : // // trivial so far.
128 : // vi_p = other.vi_p;
129 : // return *this;
130 : // }
131 :
132 0 : Bool StatWT::process(VisBuffGroup& vbg)
133 : {
134 0 : LogIO os(LogOrigin("StatWT", "process()"));
135 0 : Bool worked = true;
136 0 : uInt nvbs = vbg.nBuf();
137 0 : Int maxAnt = 0;
138 0 : Int maxNCorr = 0;
139 0 : Int maxSpw = 0; // VisBuffGroupAcc is 1 of those things that uses SpW when
140 : // it should use DDID.
141 :
142 0 : for(uInt bufnum = 0; bufnum < nvbs; ++bufnum){
143 0 : if(vbg(bufnum).numberAnt() > maxAnt) // Record maxAnt even for buffers
144 0 : maxAnt = vbg(bufnum).numberAnt(); // that won't be used in the fit.
145 0 : if(vbg(bufnum).nCorr() > maxNCorr)
146 0 : maxNCorr = vbg(bufnum).nCorr();
147 :
148 0 : if(bufnum > 0 && anyTrue(vbg(bufnum).corrType() != vbg(0).corrType())){
149 : os << LogIO::SEVERE
150 : << "statwt does not yet support combining data description IDs with different correlation setups."
151 0 : << LogIO::POST;
152 0 : return false;
153 : }
154 :
155 0 : Int spw = vbg(bufnum).spectralWindow();
156 0 : if(fitmask_p.count(spw) > 0){ // This requires fitspw to
157 : // follow the '' = nothing,
158 : // '*' = everything convention.
159 0 : if(spw > maxSpw)
160 0 : maxSpw = vbg(bufnum).spectralWindow();
161 : }
162 : }
163 :
164 0 : Cube<Bool> chanmaskedflags;
165 :
166 : // Map from hashFunction(ant1, ant2) to running number of visibilities[corr]
167 0 : std::map<uInt, Vector<uInt> > ns;
168 :
169 : // Map from hashFunction(ant1, ant2) to running mean[corr]
170 0 : std::map<uInt, Vector<Complex> > means;
171 :
172 : // Map from hashFunction(ant1, ant2) to variance[corr], initially stored as
173 : // running sums of squared differences.
174 0 : std::map<uInt, Vector<Double> > variances;
175 :
176 : // The accumulation of sums for the variances could be parallelized.
177 : // See Chan, Tony F.; Golub, Gene H.; LeVeque, Randall J. (1979), "Updating
178 : // Formulae and a Pairwise Algorithm for Computing Sample Variances.",
179 : // Technical Report STAN-CS-79-773, Department of Computer Science, Stanford
180 : // University.
181 :
182 0 : for(uInt bufnum = 0; bufnum < nvbs; ++bufnum){
183 0 : Int spw = vbg(bufnum).spectralWindow();
184 :
185 0 : if(fitmask_p.count(spw) > 0){
186 0 : VisBuffGroup::applyChanMask(chanmaskedflags, fitmask_p[spw], vbg(bufnum));
187 :
188 0 : if(!update_variances(ns, means, variances, vbg(bufnum), chanmaskedflags,
189 : maxAnt))
190 0 : return false;
191 : }
192 : }
193 0 : for(std::map<uInt, Vector<Double> >::iterator it = variances.begin();
194 0 : it != variances.end(); ++it)
195 0 : for(Int corr = 0; corr < maxNCorr; ++corr)
196 0 : it->second[corr] /= (2.*ns[it->first][corr] - 1);
197 :
198 : // TODO
199 : // if(byantenna_p){
200 : // // The formula for the variance of antenna k is
201 : // // \sigma_k^2 = \frac{1}{n - 1} \sum_{i \notequal k} \left(
202 : // // \sigma_{ik}^2 \frac{\sum_{j \notequal i,k}^{k - 1} \sigma_{jk}^2}
203 : // // {\sum_{j \notequal i,k} \sigma_{ij}^2}\right)
204 : // // where \sigma_{ij}^2 is the already calculated variance of baseline ij.
205 : //
206 : // // So, get the antenna based variances, take their sqrts \sigma_k, and
207 : // // update variances to \sigma_i \sigma_j, taking sepacs_p into account all
208 : // // along.
209 : // }
210 :
211 : //uInt oldrowsdone = rowsdone_p;
212 0 : for(uInt bufnum = 0; bufnum < nvbs; ++bufnum){
213 0 : uInt spw = vbg(bufnum).spectralWindow();
214 :
215 0 : rowsdone_p += vbg(bufnum).nRow();
216 0 : if(outspws_p.find(spw) != outspws_p.end()){
217 0 : worked &= apply_variances(vbg(bufnum), ns, variances, maxAnt);
218 : //cerr << "Wrote out row IDs " << oldrowsdone << " - " << rowsdone_p - 1 << ",";
219 : }
220 : //else
221 : // cerr << "No output for";
222 : //cerr << " spw " << spw << endl;
223 : //oldrowsdone = rowsdone_p;
224 :
225 : // Catch outvi_p up with invi_p.
226 0 : if(vbg.chunkEnd(bufnum) && outvi_p.moreChunks()){
227 0 : outvi_p.nextChunk();
228 0 : outvi_p.origin();
229 : }
230 0 : else if(outvi_p.more())
231 0 : ++outvi_p;
232 : }
233 :
234 0 : return worked;
235 0 : }
236 :
237 0 : Bool StatWT::update_variances(std::map<uInt, Vector<uInt> >& ns,
238 : std::map<uInt, Vector<Complex> >& means,
239 : std::map<uInt, Vector<Double> >& variances,
240 : const VisBuffer& vb,
241 : const Cube<Bool>& chanmaskedflags, const uInt maxAnt)
242 : {
243 0 : Cube<Complex> data(vb.dataCube(datacol_p));
244 :
245 0 : if(data.shape() != chanmaskedflags.shape())
246 0 : return false;
247 :
248 0 : Bool retval = true;
249 0 : uInt nCorr = data.shape()[0];
250 0 : uInt nChan = data.shape()[1];
251 0 : rownr_t nRows = data.shape()[2];
252 0 : Vector<uInt> unflagged(nChan);
253 0 : Vector<Int> a1(vb.antenna1());
254 0 : Vector<Int> a2(vb.antenna2());
255 :
256 0 : for(rownr_t r = 0; r < nRows; ++r){
257 0 : if(!vb.flagRow()[r]){
258 0 : uInt hr = hashFunction(a1[r], a2[r], maxAnt);
259 : // setup defaults, clear on all-flagged not needed as variances == 0 is
260 : // skipped in apply_variances
261 0 : if(!ns.count(hr)){
262 0 : ns[hr] = Vector<uInt>(nCorr, 0);
263 0 : means[hr] = Vector<Complex>(nCorr, 0);
264 0 : variances[hr] = Vector<Double>(nCorr, 0);
265 : }
266 0 : Vector<uInt> & vns = ns[hr];
267 0 : Vector<Complex> & vmeans = means[hr];
268 0 : Vector<Double> & vvariances = variances[hr];
269 :
270 0 : for(uInt corr = 0; corr < nCorr; ++corr){
271 0 : for(uInt ch = 0; ch < nChan; ++ch){
272 0 : if(!chanmaskedflags(corr, ch, r) && !vb.flagCube()(corr,ch,r)){
273 0 : Complex vis, vmoldmean, vmmean;
274 0 : ++vns[corr];
275 0 : vis = data(corr, ch, r);
276 0 : vmoldmean = vis - vmeans[corr];
277 :
278 0 : if(!dorms_p) // It's not that Complex / Int isn't defined, it's
279 : // that it is, along with Complex / Double, creating
280 : // an ambiguity.
281 0 : vmeans[corr] += vmoldmean / static_cast<Double>(vns[corr]);
282 :
283 : // This term is guaranteed to have its parts be nonnegative.
284 0 : vmmean = vis - vmeans[corr];
285 0 : vvariances[corr] += vmmean.real() * vmoldmean.real() +
286 0 : vmmean.imag() * vmoldmean.imag();
287 : }
288 : }
289 : }
290 : }
291 : }
292 0 : return retval;
293 0 : }
294 :
295 0 : Bool StatWT::apply_variances(VisBuffer& vb,
296 : std::map<uInt, Vector<uInt> >& ns,
297 : std::map<uInt, Vector<Double> >& variances,
298 : const uInt maxAnt)
299 : {
300 0 : Bool retval = true;
301 0 : IPosition shp(vb.flagCube().shape());
302 0 : uInt nCorr = shp[0];
303 0 : uInt nChan = shp[1];
304 0 : rownr_t nRows = shp[2];
305 0 : Vector<Int> a1(vb.antenna1());
306 0 : Vector<Int> a2(vb.antenna2());
307 :
308 0 : for(rownr_t r = 0; r < nRows; ++r){
309 0 : uInt hr = hashFunction(a1[r], a2[r], maxAnt);
310 0 : Bool unflagged = false;
311 0 : Bool havevar = ns.count(hr) > 0;
312 :
313 0 : for(uInt corr = 0; corr < nCorr; ++corr){
314 0 : if(havevar &&
315 0 : (ns[hr][corr] >= minsamp_p) &&
316 0 : (0.0 < variances[hr][corr])){ // For some reason emacs likes 0 < v,
317 : // but not v > 0.
318 0 : Double var = variances[hr][corr];
319 :
320 0 : unflagged = true;
321 0 : vb.sigmaMat()(corr, r) = sqrt(var);
322 0 : vb.weightMat()(corr, r) = 1.0 / var;
323 : }
324 : else{
325 0 : vb.sigmaMat()(corr, r) = -1.0;
326 0 : vb.weightMat()(corr, r) = 0.0;
327 0 : for(uInt ch = 0; ch < nChan; ++ch){
328 0 : vb.flagCube()(corr, ch, r) = true;
329 : }
330 : }
331 0 : if(!unflagged)
332 0 : vb.flagRow()[r] = true;
333 : }
334 : }
335 :
336 : // argh
337 : // outvi_p.setFlagCube(vb.flagCube());
338 0 : outvi_p.setFlag(vb.flag());
339 :
340 0 : outvi_p.setSigmaMat(vb.sigmaMat());
341 0 : outvi_p.setWeightMat(vb.weightMat());
342 0 : return retval;
343 0 : }
344 :
345 : using namespace casacore;
346 : } // end namespace casa
|
