1 | //#---------------------------------------------------------------------------
|
---|
2 | //# SDMath.cc: A collection of single dish mathematical operations
|
---|
3 | //#---------------------------------------------------------------------------
|
---|
4 | //# Copyright (C) 2004
|
---|
5 | //# ATNF
|
---|
6 | //#
|
---|
7 | //# This program is free software; you can redistribute it and/or modify it
|
---|
8 | //# under the terms of the GNU General Public License as published by the Free
|
---|
9 | //# Software Foundation; either version 2 of the License, or (at your option)
|
---|
10 | //# any later version.
|
---|
11 | //#
|
---|
12 | //# This program is distributed in the hope that it will be useful, but
|
---|
13 | //# WITHOUT ANY WARRANTY; without even the implied warranty of
|
---|
14 | //# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General
|
---|
15 | //# Public License for more details.
|
---|
16 | //#
|
---|
17 | //# You should have received a copy of the GNU General Public License along
|
---|
18 | //# with this program; if not, write to the Free Software Foundation, Inc.,
|
---|
19 | //# 675 Massachusetts Ave, Cambridge, MA 02139, USA.
|
---|
20 | //#
|
---|
21 | //# Correspondence concerning this software should be addressed as follows:
|
---|
22 | //# Internet email: Malte.Marquarding@csiro.au
|
---|
23 | //# Postal address: Malte Marquarding,
|
---|
24 | //# Australia Telescope National Facility,
|
---|
25 | //# P.O. Box 76,
|
---|
26 | //# Epping, NSW, 2121,
|
---|
27 | //# AUSTRALIA
|
---|
28 | //#
|
---|
29 | //# $Id:
|
---|
30 | //#---------------------------------------------------------------------------
|
---|
31 | #include <vector>
|
---|
32 |
|
---|
33 | #include <casa/aips.h>
|
---|
34 | #include <casa/BasicSL/String.h>
|
---|
35 | #include <casa/Arrays/IPosition.h>
|
---|
36 | #include <casa/Arrays/Array.h>
|
---|
37 | #include <casa/Arrays/ArrayIter.h>
|
---|
38 | #include <casa/Arrays/VectorIter.h>
|
---|
39 | #include <casa/Arrays/ArrayMath.h>
|
---|
40 | #include <casa/Arrays/ArrayLogical.h>
|
---|
41 | #include <casa/Arrays/MaskedArray.h>
|
---|
42 | #include <casa/Arrays/MaskArrMath.h>
|
---|
43 | #include <casa/Arrays/MaskArrLogi.h>
|
---|
44 | #include <casa/BasicMath/Math.h>
|
---|
45 | #include <casa/Containers/Block.h>
|
---|
46 | #include <casa/Exceptions.h>
|
---|
47 | #include <casa/Quanta/Quantum.h>
|
---|
48 | #include <casa/Quanta/Unit.h>
|
---|
49 | #include <casa/Quanta/MVEpoch.h>
|
---|
50 | #include <casa/Quanta/QC.h>
|
---|
51 | #include <casa/Quanta/MVTime.h>
|
---|
52 | #include <casa/Utilities/Assert.h>
|
---|
53 |
|
---|
54 | #include <coordinates/Coordinates/SpectralCoordinate.h>
|
---|
55 | #include <coordinates/Coordinates/CoordinateSystem.h>
|
---|
56 | #include <coordinates/Coordinates/CoordinateUtil.h>
|
---|
57 | #include <coordinates/Coordinates/VelocityAligner.h>
|
---|
58 |
|
---|
59 | #include <lattices/Lattices/LatticeUtilities.h>
|
---|
60 | #include <lattices/Lattices/RebinLattice.h>
|
---|
61 |
|
---|
62 | #include <measures/Measures/MEpoch.h>
|
---|
63 | #include <measures/Measures/MDirection.h>
|
---|
64 | #include <measures/Measures/MPosition.h>
|
---|
65 |
|
---|
66 | #include <scimath/Mathematics/VectorKernel.h>
|
---|
67 | #include <scimath/Mathematics/Convolver.h>
|
---|
68 | #include <scimath/Mathematics/InterpolateArray1D.h>
|
---|
69 | #include <scimath/Functionals/Polynomial.h>
|
---|
70 |
|
---|
71 | #include <tables/Tables/Table.h>
|
---|
72 | #include <tables/Tables/ScalarColumn.h>
|
---|
73 | #include <tables/Tables/ArrayColumn.h>
|
---|
74 | #include <tables/Tables/ReadAsciiTable.h>
|
---|
75 |
|
---|
76 | #include "MathUtils.h"
|
---|
77 | #include "SDDefs.h"
|
---|
78 | #include "SDContainer.h"
|
---|
79 | #include "SDMemTable.h"
|
---|
80 |
|
---|
81 | #include "SDMath.h"
|
---|
82 |
|
---|
83 | using namespace casa;
|
---|
84 | using namespace asap;
|
---|
85 |
|
---|
86 |
|
---|
87 | SDMath::SDMath()
|
---|
88 | {;}
|
---|
89 |
|
---|
90 | SDMath::SDMath(const SDMath& other)
|
---|
91 | {
|
---|
92 |
|
---|
93 | // No state
|
---|
94 |
|
---|
95 | }
|
---|
96 |
|
---|
97 | SDMath& SDMath::operator=(const SDMath& other)
|
---|
98 | {
|
---|
99 | if (this != &other) {
|
---|
100 | // No state
|
---|
101 | }
|
---|
102 | return *this;
|
---|
103 | }
|
---|
104 |
|
---|
105 | SDMath::~SDMath()
|
---|
106 | {;}
|
---|
107 |
|
---|
108 |
|
---|
109 |
|
---|
110 | SDMemTable* SDMath::velocityAlignment (const SDMemTable& in, const String& refTime) const
|
---|
111 | {
|
---|
112 |
|
---|
113 | // Get velocity/frame info from Table
|
---|
114 |
|
---|
115 | std::vector<std::string> info = in.getCoordInfo();
|
---|
116 | String velUnit(info[0]);
|
---|
117 | if (velUnit.length()==0) {
|
---|
118 | throw(AipsError("You have not set a velocity abcissa unit - use function set_unit"));
|
---|
119 | } else {
|
---|
120 | Unit velUnitU(velUnit);
|
---|
121 | if (velUnitU!=Unit(String("m/s"))) {
|
---|
122 | throw(AipsError("Specified abcissa unit is not consistent with km/s - use function set_unit"));
|
---|
123 | }
|
---|
124 | }
|
---|
125 | //
|
---|
126 | String dopplerStr(info[2]);
|
---|
127 | String velSystemStr(info[1]);
|
---|
128 | String velBaseSystemStr(info[3]);
|
---|
129 | if (velBaseSystemStr==velSystemStr) {
|
---|
130 | throw(AipsError("You have not set a velocity frame different from the initial - use function set_freqframe"));
|
---|
131 | }
|
---|
132 | //
|
---|
133 | MFrequency::Types velSystem;
|
---|
134 | MFrequency::getType(velSystem, velSystemStr);
|
---|
135 | MDoppler::Types doppler;
|
---|
136 | MDoppler::getType(doppler, dopplerStr);
|
---|
137 |
|
---|
138 | // Decide on alignment Epoch
|
---|
139 |
|
---|
140 | // Do it
|
---|
141 |
|
---|
142 | return velocityAlign (in, velSystem, velUnit, doppler, refTime);
|
---|
143 | }
|
---|
144 |
|
---|
145 |
|
---|
146 |
|
---|
147 | CountedPtr<SDMemTable> SDMath::average(const Block<CountedPtr<SDMemTable> >& in,
|
---|
148 | const Vector<Bool>& mask, Bool scanAv,
|
---|
149 | const String& weightStr, Bool alignVelocity) const
|
---|
150 | //
|
---|
151 | // Weighted averaging of spectra from one or more Tables.
|
---|
152 | //
|
---|
153 | {
|
---|
154 |
|
---|
155 | // Convert weight type
|
---|
156 |
|
---|
157 | WeightType wtType = NONE;
|
---|
158 | convertWeightString(wtType, weightStr);
|
---|
159 |
|
---|
160 | // Create output Table by cloning from the first table
|
---|
161 |
|
---|
162 | SDMemTable* pTabOut = new SDMemTable(*in[0],True);
|
---|
163 |
|
---|
164 | // Setup
|
---|
165 |
|
---|
166 | IPosition shp = in[0]->rowAsMaskedArray(0).shape(); // Must not change
|
---|
167 | Array<Float> arr(shp);
|
---|
168 | Array<Bool> barr(shp);
|
---|
169 | const Bool useMask = (mask.nelements() == shp(asap::ChanAxis));
|
---|
170 |
|
---|
171 | // Columns from Tables
|
---|
172 |
|
---|
173 | ROArrayColumn<Float> tSysCol;
|
---|
174 | ROScalarColumn<Double> mjdCol;
|
---|
175 | ROScalarColumn<String> srcNameCol;
|
---|
176 | ROScalarColumn<Double> intCol;
|
---|
177 | ROArrayColumn<uInt> fqIDCol;
|
---|
178 |
|
---|
179 | // Create accumulation MaskedArray. We accumulate for each channel,if,pol,beam
|
---|
180 | // Note that the mask of the accumulation array will ALWAYS remain ALL True.
|
---|
181 | // The MA is only used so that when data which is masked Bad is added to it,
|
---|
182 | // that data does not contribute.
|
---|
183 |
|
---|
184 | Array<Float> zero(shp);
|
---|
185 | zero=0.0;
|
---|
186 | Array<Bool> good(shp);
|
---|
187 | good = True;
|
---|
188 | MaskedArray<Float> sum(zero,good);
|
---|
189 |
|
---|
190 | // Counter arrays
|
---|
191 |
|
---|
192 | Array<Float> nPts(shp); // Number of points
|
---|
193 | nPts = 0.0;
|
---|
194 | Array<Float> nInc(shp); // Increment
|
---|
195 | nInc = 1.0;
|
---|
196 |
|
---|
197 | // Create accumulation Array for variance. We accumulate for
|
---|
198 | // each if,pol,beam, but average over channel. So we need
|
---|
199 | // a shape with one less axis dropping channels.
|
---|
200 |
|
---|
201 | const uInt nAxesSub = shp.nelements() - 1;
|
---|
202 | IPosition shp2(nAxesSub);
|
---|
203 | for (uInt i=0,j=0; i<(nAxesSub+1); i++) {
|
---|
204 | if (i!=asap::ChanAxis) {
|
---|
205 | shp2(j) = shp(i);
|
---|
206 | j++;
|
---|
207 | }
|
---|
208 | }
|
---|
209 | Array<Float> sumSq(shp2);
|
---|
210 | sumSq = 0.0;
|
---|
211 | IPosition pos2(nAxesSub,0); // For indexing
|
---|
212 |
|
---|
213 | // Time-related accumulators
|
---|
214 |
|
---|
215 | Double time;
|
---|
216 | Double timeSum = 0.0;
|
---|
217 | Double intSum = 0.0;
|
---|
218 | Double interval = 0.0;
|
---|
219 |
|
---|
220 | // To get the right shape for the Tsys accumulator we need to
|
---|
221 | // access a column from the first table. The shape of this
|
---|
222 | // array must not change
|
---|
223 |
|
---|
224 | Array<Float> tSysSum;
|
---|
225 | {
|
---|
226 | const Table& tabIn = in[0]->table();
|
---|
227 | tSysCol.attach(tabIn,"TSYS");
|
---|
228 | tSysSum.resize(tSysCol.shape(0));
|
---|
229 | }
|
---|
230 | tSysSum =0.0;
|
---|
231 | Array<Float> tSys;
|
---|
232 |
|
---|
233 | // Scan and row tracking
|
---|
234 |
|
---|
235 | Int oldScanID = 0;
|
---|
236 | Int outScanID = 0;
|
---|
237 | Int scanID = 0;
|
---|
238 | Int rowStart = 0;
|
---|
239 | Int nAccum = 0;
|
---|
240 | Int tableStart = 0;
|
---|
241 |
|
---|
242 | // Source and FreqID
|
---|
243 |
|
---|
244 | String sourceName, oldSourceName, sourceNameStart;
|
---|
245 | Vector<uInt> freqID, freqIDStart, oldFreqID;
|
---|
246 |
|
---|
247 | // Loop over tables
|
---|
248 |
|
---|
249 | Float fac = 1.0;
|
---|
250 | const uInt nTables = in.nelements();
|
---|
251 | for (uInt iTab=0; iTab<nTables; iTab++) {
|
---|
252 |
|
---|
253 | // Should check that the frequency tables don't change if doing VelocityAlignment
|
---|
254 |
|
---|
255 | // Attach columns to Table
|
---|
256 |
|
---|
257 | const Table& tabIn = in[iTab]->table();
|
---|
258 | tSysCol.attach(tabIn, "TSYS");
|
---|
259 | mjdCol.attach(tabIn, "TIME");
|
---|
260 | srcNameCol.attach(tabIn, "SRCNAME");
|
---|
261 | intCol.attach(tabIn, "INTERVAL");
|
---|
262 | fqIDCol.attach(tabIn, "FREQID");
|
---|
263 |
|
---|
264 | // Loop over rows in Table
|
---|
265 |
|
---|
266 | const uInt nRows = in[iTab]->nRow();
|
---|
267 | for (uInt iRow=0; iRow<nRows; iRow++) {
|
---|
268 |
|
---|
269 | // Check conformance
|
---|
270 |
|
---|
271 | IPosition shp2 = in[iTab]->rowAsMaskedArray(iRow).shape();
|
---|
272 | if (!shp.isEqual(shp2)) {
|
---|
273 | throw (AipsError("Shapes for all rows must be the same"));
|
---|
274 | }
|
---|
275 |
|
---|
276 | // If we are not doing scan averages, make checks for source and
|
---|
277 | // frequency setup and warn if averaging across them
|
---|
278 |
|
---|
279 | // Get copy of Scan Container for this row
|
---|
280 |
|
---|
281 | SDContainer sc = in[iTab]->getSDContainer(iRow);
|
---|
282 | scanID = sc.scanid;
|
---|
283 |
|
---|
284 | // Get quantities from columns
|
---|
285 |
|
---|
286 | srcNameCol.getScalar(iRow, sourceName);
|
---|
287 | mjdCol.get(iRow, time);
|
---|
288 | tSysCol.get(iRow, tSys);
|
---|
289 | intCol.get(iRow, interval);
|
---|
290 | fqIDCol.get(iRow, freqID);
|
---|
291 |
|
---|
292 | // Initialize first source and freqID
|
---|
293 |
|
---|
294 | if (iRow==0 && iTab==0) {
|
---|
295 | sourceNameStart = sourceName;
|
---|
296 | freqIDStart = freqID;
|
---|
297 | }
|
---|
298 |
|
---|
299 | // If we are doing scan averages, see if we are at the end of an
|
---|
300 | // accumulation period (scan). We must check soutce names too,
|
---|
301 | // since we might have two tables with one scan each but different
|
---|
302 | // source names; we shouldn't average different sources together
|
---|
303 |
|
---|
304 | if (scanAv && ( (scanID != oldScanID) ||
|
---|
305 | (iRow==0 && iTab>0 && sourceName!=oldSourceName))) {
|
---|
306 |
|
---|
307 | // Normalize data in 'sum' accumulation array according to weighting scheme
|
---|
308 |
|
---|
309 | normalize(sum, sumSq, nPts, wtType, asap::ChanAxis, nAxesSub);
|
---|
310 |
|
---|
311 | // Fill scan container. The source and freqID come from the
|
---|
312 | // first row of the first table that went into this average (
|
---|
313 | // should be the same for all rows in the scan average)
|
---|
314 |
|
---|
315 | Float nR(nAccum);
|
---|
316 | fillSDC(sc, sum.getMask(), sum.getArray(), tSysSum/nR, outScanID,
|
---|
317 | timeSum/nR, intSum, sourceNameStart, freqIDStart);
|
---|
318 |
|
---|
319 | // Write container out to Table
|
---|
320 |
|
---|
321 | pTabOut->putSDContainer(sc);
|
---|
322 |
|
---|
323 | // Reset accumulators
|
---|
324 |
|
---|
325 | sum = 0.0;
|
---|
326 | sumSq = 0.0;
|
---|
327 | nAccum = 0;
|
---|
328 | //
|
---|
329 | tSysSum =0.0;
|
---|
330 | timeSum = 0.0;
|
---|
331 | intSum = 0.0;
|
---|
332 | nPts = 0.0;
|
---|
333 |
|
---|
334 | // Increment
|
---|
335 |
|
---|
336 | rowStart = iRow; // First row for next accumulation
|
---|
337 | tableStart = iTab; // First table for next accumulation
|
---|
338 | sourceNameStart = sourceName; // First source name for next accumulation
|
---|
339 | freqIDStart = freqID; // First FreqID for next accumulation
|
---|
340 | //
|
---|
341 | oldScanID = scanID;
|
---|
342 | outScanID += 1; // Scan ID for next accumulation period
|
---|
343 | }
|
---|
344 |
|
---|
345 | // Accumulate
|
---|
346 |
|
---|
347 | accumulate(timeSum, intSum, nAccum, sum, sumSq, nPts, tSysSum,
|
---|
348 | tSys, nInc, mask, time, interval, in, iTab, iRow, asap::ChanAxis,
|
---|
349 | nAxesSub, useMask, wtType);
|
---|
350 | //
|
---|
351 | oldSourceName = sourceName;
|
---|
352 | oldFreqID = freqID;
|
---|
353 | }
|
---|
354 | }
|
---|
355 |
|
---|
356 | // OK at this point we have accumulation data which is either
|
---|
357 | // - accumulated from all tables into one row
|
---|
358 | // or
|
---|
359 | // - accumulated from the last scan average
|
---|
360 | //
|
---|
361 | // Normalize data in 'sum' accumulation array according to weighting scheme
|
---|
362 | normalize(sum, sumSq, nPts, wtType, asap::ChanAxis, nAxesSub);
|
---|
363 |
|
---|
364 | // Create and fill container. The container we clone will be from
|
---|
365 | // the last Table and the first row that went into the current
|
---|
366 | // accumulation. It probably doesn't matter that much really...
|
---|
367 |
|
---|
368 | Float nR(nAccum);
|
---|
369 | SDContainer sc = in[tableStart]->getSDContainer(rowStart);
|
---|
370 | fillSDC(sc, sum.getMask(), sum.getArray(), tSysSum/nR, outScanID,
|
---|
371 | timeSum/nR, intSum, sourceNameStart, freqIDStart);
|
---|
372 | pTabOut->putSDContainer(sc);
|
---|
373 | //
|
---|
374 | return CountedPtr<SDMemTable>(pTabOut);
|
---|
375 | }
|
---|
376 |
|
---|
377 |
|
---|
378 |
|
---|
379 | CountedPtr<SDMemTable> SDMath::binaryOperate (const CountedPtr<SDMemTable>& left,
|
---|
380 | const CountedPtr<SDMemTable>& right,
|
---|
381 | const String& op, Bool preserve) const
|
---|
382 | {
|
---|
383 |
|
---|
384 | // Check operator
|
---|
385 |
|
---|
386 | String op2(op);
|
---|
387 | op2.upcase();
|
---|
388 | uInt what = 0;
|
---|
389 | if (op2=="ADD") {
|
---|
390 | what = 0;
|
---|
391 | } else if (op2=="SUB") {
|
---|
392 | what = 1;
|
---|
393 | } else if (op2=="MUL") {
|
---|
394 | what = 2;
|
---|
395 | } else if (op2=="DIV") {
|
---|
396 | what = 3;
|
---|
397 | } else if (op2=="QUOTIENT") {
|
---|
398 | what = 4;
|
---|
399 | } else {
|
---|
400 | throw( AipsError("Unrecognized operation"));
|
---|
401 | }
|
---|
402 |
|
---|
403 | // Check rows
|
---|
404 |
|
---|
405 | const uInt nRowLeft = left->nRow();
|
---|
406 | const uInt nRowRight = right->nRow();
|
---|
407 | Bool ok = (nRowRight==1&&nRowLeft>0) ||
|
---|
408 | (nRowRight>=1&&nRowLeft==nRowRight);
|
---|
409 | if (!ok) {
|
---|
410 | throw (AipsError("The right Scan Table can have one row or the same number of rows as the left Scan Table"));
|
---|
411 | }
|
---|
412 |
|
---|
413 | // Input Tables
|
---|
414 |
|
---|
415 | const Table& tLeft = left->table();
|
---|
416 | const Table& tRight = right->table();
|
---|
417 |
|
---|
418 | // TSys columns
|
---|
419 |
|
---|
420 | ROArrayColumn<Float> tSysLeft(tLeft, "TSYS");
|
---|
421 | ROArrayColumn<Float> tSysRight(tRight, "TSYS");
|
---|
422 |
|
---|
423 | // First row for right
|
---|
424 |
|
---|
425 | Array<Float> tSysLeftArr, tSysRightArr;
|
---|
426 | tSysRight.get(0, tSysRightArr);
|
---|
427 | MaskedArray<Float>* pMRight = new MaskedArray<Float>(right->rowAsMaskedArray(0));
|
---|
428 | IPosition shpRight = pMRight->shape();
|
---|
429 |
|
---|
430 | // Output Table cloned from left
|
---|
431 |
|
---|
432 | SDMemTable* pTabOut = new SDMemTable(*left, True);
|
---|
433 |
|
---|
434 | // Loop over rows
|
---|
435 |
|
---|
436 | for (uInt i=0; i<nRowLeft; i++) {
|
---|
437 |
|
---|
438 | // Get data
|
---|
439 |
|
---|
440 | MaskedArray<Float> mLeft(left->rowAsMaskedArray(i));
|
---|
441 | IPosition shpLeft = mLeft.shape();
|
---|
442 | tSysLeft.get(i, tSysLeftArr);
|
---|
443 | //
|
---|
444 | if (nRowRight>1) {
|
---|
445 | delete pMRight;
|
---|
446 | pMRight = new MaskedArray<Float>(right->rowAsMaskedArray(i));
|
---|
447 | shpRight = pMRight->shape();
|
---|
448 | tSysRight.get(i, tSysRightArr);
|
---|
449 | }
|
---|
450 | //
|
---|
451 | if (!shpRight.isEqual(shpLeft)) {
|
---|
452 | throw(AipsError("left and right scan tables are not conformant"));
|
---|
453 | }
|
---|
454 | if (!tSysRightArr.shape().isEqual(tSysRightArr.shape())) {
|
---|
455 | throw(AipsError("left and right Tsys data are not conformant"));
|
---|
456 | }
|
---|
457 | if (!shpRight.isEqual(tSysRightArr.shape())) {
|
---|
458 | throw(AipsError("left and right scan tables are not conformant"));
|
---|
459 | }
|
---|
460 |
|
---|
461 | // Make container
|
---|
462 |
|
---|
463 | SDContainer sc = left->getSDContainer(i);
|
---|
464 |
|
---|
465 | // Operate on data and TSys
|
---|
466 |
|
---|
467 | if (what==0) {
|
---|
468 | MaskedArray<Float> tmp = mLeft + *pMRight;
|
---|
469 | putDataInSDC(sc, tmp.getArray(), tmp.getMask());
|
---|
470 | sc.putTsys(tSysLeftArr+tSysRightArr);
|
---|
471 | } else if (what==1) {
|
---|
472 | MaskedArray<Float> tmp = mLeft - *pMRight;
|
---|
473 | putDataInSDC(sc, tmp.getArray(), tmp.getMask());
|
---|
474 | sc.putTsys(tSysLeftArr-tSysRightArr);
|
---|
475 | } else if (what==2) {
|
---|
476 | MaskedArray<Float> tmp = mLeft * *pMRight;
|
---|
477 | putDataInSDC(sc, tmp.getArray(), tmp.getMask());
|
---|
478 | sc.putTsys(tSysLeftArr*tSysRightArr);
|
---|
479 | } else if (what==3) {
|
---|
480 | MaskedArray<Float> tmp = mLeft / *pMRight;
|
---|
481 | putDataInSDC(sc, tmp.getArray(), tmp.getMask());
|
---|
482 | sc.putTsys(tSysLeftArr/tSysRightArr);
|
---|
483 | } else if (what==4) {
|
---|
484 | if (preserve) {
|
---|
485 | MaskedArray<Float> tmp = (tSysRightArr * mLeft / *pMRight) - tSysRightArr;
|
---|
486 | putDataInSDC(sc, tmp.getArray(), tmp.getMask());
|
---|
487 | } else {
|
---|
488 | MaskedArray<Float> tmp = (tSysRightArr * mLeft / *pMRight) - tSysLeftArr;
|
---|
489 | putDataInSDC(sc, tmp.getArray(), tmp.getMask());
|
---|
490 | }
|
---|
491 | sc.putTsys(tSysRightArr);
|
---|
492 | }
|
---|
493 |
|
---|
494 | // Put new row in output Table
|
---|
495 |
|
---|
496 | pTabOut->putSDContainer(sc);
|
---|
497 | }
|
---|
498 | if (pMRight) delete pMRight;
|
---|
499 | //
|
---|
500 | return CountedPtr<SDMemTable>(pTabOut);
|
---|
501 | }
|
---|
502 |
|
---|
503 |
|
---|
504 |
|
---|
505 | std::vector<float> SDMath::statistic(const CountedPtr<SDMemTable>& in,
|
---|
506 | const Vector<Bool>& mask,
|
---|
507 | const String& which, Int row) const
|
---|
508 | //
|
---|
509 | // Perhaps iteration over pol/beam/if should be in here
|
---|
510 | // and inside the nrow iteration ?
|
---|
511 | //
|
---|
512 | {
|
---|
513 | const uInt nRow = in->nRow();
|
---|
514 |
|
---|
515 | // Specify cursor location
|
---|
516 |
|
---|
517 | IPosition start, end;
|
---|
518 | getCursorLocation(start, end, *in);
|
---|
519 |
|
---|
520 | // Loop over rows
|
---|
521 |
|
---|
522 | const uInt nEl = mask.nelements();
|
---|
523 | uInt iStart = 0;
|
---|
524 | uInt iEnd = in->nRow()-1;
|
---|
525 | //
|
---|
526 | if (row>=0) {
|
---|
527 | iStart = row;
|
---|
528 | iEnd = row;
|
---|
529 | }
|
---|
530 | //
|
---|
531 | std::vector<float> result(iEnd-iStart+1);
|
---|
532 | for (uInt ii=iStart; ii <= iEnd; ++ii) {
|
---|
533 |
|
---|
534 | // Get row and deconstruct
|
---|
535 |
|
---|
536 | MaskedArray<Float> marr(in->rowAsMaskedArray(ii));
|
---|
537 | Array<Float> arr = marr.getArray();
|
---|
538 | Array<Bool> barr = marr.getMask();
|
---|
539 |
|
---|
540 | // Access desired piece of data
|
---|
541 |
|
---|
542 | Array<Float> v((arr(start,end)).nonDegenerate());
|
---|
543 | Array<Bool> m((barr(start,end)).nonDegenerate());
|
---|
544 |
|
---|
545 | // Apply OTF mask
|
---|
546 |
|
---|
547 | MaskedArray<Float> tmp;
|
---|
548 | if (m.nelements()==nEl) {
|
---|
549 | tmp.setData(v,m&&mask);
|
---|
550 | } else {
|
---|
551 | tmp.setData(v,m);
|
---|
552 | }
|
---|
553 |
|
---|
554 | // Get statistic
|
---|
555 |
|
---|
556 | result[ii-iStart] = mathutil::statistics(which, tmp);
|
---|
557 | }
|
---|
558 | //
|
---|
559 | return result;
|
---|
560 | }
|
---|
561 |
|
---|
562 |
|
---|
563 | SDMemTable* SDMath::bin(const SDMemTable& in, Int width) const
|
---|
564 | {
|
---|
565 | SDHeader sh = in.getSDHeader();
|
---|
566 | SDMemTable* pTabOut = new SDMemTable(in, True);
|
---|
567 |
|
---|
568 | // Bin up SpectralCoordinates
|
---|
569 |
|
---|
570 | IPosition factors(1);
|
---|
571 | factors(0) = width;
|
---|
572 | for (uInt j=0; j<in.nCoordinates(); ++j) {
|
---|
573 | CoordinateSystem cSys;
|
---|
574 | cSys.addCoordinate(in.getCoordinate(j));
|
---|
575 | CoordinateSystem cSysBin =
|
---|
576 | CoordinateUtil::makeBinnedCoordinateSystem(factors, cSys, False);
|
---|
577 | //
|
---|
578 | SpectralCoordinate sCBin = cSysBin.spectralCoordinate(0);
|
---|
579 | pTabOut->setCoordinate(sCBin, j);
|
---|
580 | }
|
---|
581 |
|
---|
582 | // Use RebinLattice to find shape
|
---|
583 |
|
---|
584 | IPosition shapeIn(1,sh.nchan);
|
---|
585 | IPosition shapeOut = RebinLattice<Float>::rebinShape(shapeIn, factors);
|
---|
586 | sh.nchan = shapeOut(0);
|
---|
587 | pTabOut->putSDHeader(sh);
|
---|
588 |
|
---|
589 |
|
---|
590 | // Loop over rows and bin along channel axis
|
---|
591 |
|
---|
592 | for (uInt i=0; i < in.nRow(); ++i) {
|
---|
593 | SDContainer sc = in.getSDContainer(i);
|
---|
594 | //
|
---|
595 | Array<Float> tSys(sc.getTsys()); // Get it out before sc changes shape
|
---|
596 |
|
---|
597 | // Bin up spectrum
|
---|
598 |
|
---|
599 | MaskedArray<Float> marr(in.rowAsMaskedArray(i));
|
---|
600 | MaskedArray<Float> marrout;
|
---|
601 | LatticeUtilities::bin(marrout, marr, asap::ChanAxis, width);
|
---|
602 |
|
---|
603 | // Put back the binned data and flags
|
---|
604 |
|
---|
605 | IPosition ip2 = marrout.shape();
|
---|
606 | sc.resize(ip2);
|
---|
607 | //
|
---|
608 | putDataInSDC(sc, marrout.getArray(), marrout.getMask());
|
---|
609 |
|
---|
610 | // Bin up Tsys.
|
---|
611 |
|
---|
612 | Array<Bool> allGood(tSys.shape(),True);
|
---|
613 | MaskedArray<Float> tSysIn(tSys, allGood, True);
|
---|
614 | //
|
---|
615 | MaskedArray<Float> tSysOut;
|
---|
616 | LatticeUtilities::bin(tSysOut, tSysIn, asap::ChanAxis, width);
|
---|
617 | sc.putTsys(tSysOut.getArray());
|
---|
618 | //
|
---|
619 | pTabOut->putSDContainer(sc);
|
---|
620 | }
|
---|
621 | return pTabOut;
|
---|
622 | }
|
---|
623 |
|
---|
624 | SDMemTable* SDMath::unaryOperate(const SDMemTable& in, Float val, Bool doAll,
|
---|
625 | uInt what) const
|
---|
626 | //
|
---|
627 | // what = 0 Multiply
|
---|
628 | // 1 Add
|
---|
629 | {
|
---|
630 | SDMemTable* pOut = new SDMemTable(in,False);
|
---|
631 | const Table& tOut = pOut->table();
|
---|
632 | ArrayColumn<Float> spec(tOut,"SPECTRA");
|
---|
633 | //
|
---|
634 | if (doAll) {
|
---|
635 | for (uInt i=0; i < tOut.nrow(); i++) {
|
---|
636 | MaskedArray<Float> dataIn(pOut->rowAsMaskedArray(i));
|
---|
637 | //
|
---|
638 | if (what==0) {
|
---|
639 | dataIn *= val;
|
---|
640 | } else if (what==1) {
|
---|
641 | dataIn += val;
|
---|
642 | }
|
---|
643 | //
|
---|
644 | spec.put(i, dataIn.getArray());
|
---|
645 | }
|
---|
646 | } else {
|
---|
647 |
|
---|
648 | // Get cursor location
|
---|
649 |
|
---|
650 | IPosition start, end;
|
---|
651 | getCursorLocation(start, end, in);
|
---|
652 | //
|
---|
653 | for (uInt i=0; i < tOut.nrow(); i++) {
|
---|
654 | MaskedArray<Float> dataIn(pOut->rowAsMaskedArray(i));
|
---|
655 | MaskedArray<Float> dataIn2 = dataIn(start,end); // Reference
|
---|
656 | //
|
---|
657 | if (what==0) {
|
---|
658 | dataIn2 *= val;
|
---|
659 | } else if (what==1) {
|
---|
660 | dataIn2 += val;
|
---|
661 | }
|
---|
662 | //
|
---|
663 | spec.put(i, dataIn.getArray());
|
---|
664 | }
|
---|
665 | }
|
---|
666 | //
|
---|
667 | return pOut;
|
---|
668 | }
|
---|
669 |
|
---|
670 |
|
---|
671 |
|
---|
672 | SDMemTable* SDMath::averagePol(const SDMemTable& in, const Vector<Bool>& mask) const
|
---|
673 | //
|
---|
674 | // Average all polarizations together, weighted by variance
|
---|
675 | //
|
---|
676 | {
|
---|
677 | // WeightType wtType = NONE;
|
---|
678 | // convertWeightString(wtType, weight);
|
---|
679 |
|
---|
680 | const uInt nRows = in.nRow();
|
---|
681 |
|
---|
682 | // Create output Table and reshape number of polarizations
|
---|
683 |
|
---|
684 | Bool clear=True;
|
---|
685 | SDMemTable* pTabOut = new SDMemTable(in, clear);
|
---|
686 | SDHeader header = pTabOut->getSDHeader();
|
---|
687 | header.npol = 1;
|
---|
688 | pTabOut->putSDHeader(header);
|
---|
689 |
|
---|
690 | // Shape of input and output data
|
---|
691 |
|
---|
692 | const IPosition& shapeIn = in.rowAsMaskedArray(0u, False).shape();
|
---|
693 | IPosition shapeOut(shapeIn);
|
---|
694 | shapeOut(asap::PolAxis) = 1; // Average all polarizations
|
---|
695 | //
|
---|
696 | const uInt nChan = shapeIn(asap::ChanAxis);
|
---|
697 | const IPosition vecShapeOut(4,1,1,1,nChan); // A multi-dim form of a Vector shape
|
---|
698 | IPosition start(4), end(4);
|
---|
699 |
|
---|
700 | // Output arrays
|
---|
701 |
|
---|
702 | Array<Float> outData(shapeOut, 0.0);
|
---|
703 | Array<Bool> outMask(shapeOut, True);
|
---|
704 | const IPosition axes(2, asap::PolAxis, asap::ChanAxis); // pol-channel plane
|
---|
705 | //
|
---|
706 | const Bool useMask = (mask.nelements() == shapeIn(asap::ChanAxis));
|
---|
707 |
|
---|
708 | // Loop over rows
|
---|
709 |
|
---|
710 | for (uInt iRow=0; iRow<nRows; iRow++) {
|
---|
711 |
|
---|
712 | // Get data for this row
|
---|
713 |
|
---|
714 | MaskedArray<Float> marr(in.rowAsMaskedArray(iRow));
|
---|
715 | Array<Float>& arr = marr.getRWArray();
|
---|
716 | const Array<Bool>& barr = marr.getMask();
|
---|
717 |
|
---|
718 | // Make iterators to iterate by pol-channel planes
|
---|
719 |
|
---|
720 | ReadOnlyArrayIterator<Float> itDataPlane(arr, axes);
|
---|
721 | ReadOnlyArrayIterator<Bool> itMaskPlane(barr, axes);
|
---|
722 |
|
---|
723 | // Accumulations
|
---|
724 |
|
---|
725 | Float fac = 1.0;
|
---|
726 | Vector<Float> vecSum(nChan,0.0);
|
---|
727 |
|
---|
728 | // Iterate through data by pol-channel planes
|
---|
729 |
|
---|
730 | while (!itDataPlane.pastEnd()) {
|
---|
731 |
|
---|
732 | // Iterate through plane by polarization and accumulate Vectors
|
---|
733 |
|
---|
734 | Vector<Float> t1(nChan); t1 = 0.0;
|
---|
735 | Vector<Bool> t2(nChan); t2 = True;
|
---|
736 | MaskedArray<Float> vecSum(t1,t2);
|
---|
737 | Float varSum = 0.0;
|
---|
738 | {
|
---|
739 | ReadOnlyVectorIterator<Float> itDataVec(itDataPlane.array(), 1);
|
---|
740 | ReadOnlyVectorIterator<Bool> itMaskVec(itMaskPlane.array(), 1);
|
---|
741 | while (!itDataVec.pastEnd()) {
|
---|
742 |
|
---|
743 | // Create MA of data & mask (optionally including OTF mask) and get variance
|
---|
744 |
|
---|
745 | if (useMask) {
|
---|
746 | const MaskedArray<Float> spec(itDataVec.vector(),mask&&itMaskVec.vector());
|
---|
747 | fac = 1.0 / variance(spec);
|
---|
748 | } else {
|
---|
749 | const MaskedArray<Float> spec(itDataVec.vector(),itMaskVec.vector());
|
---|
750 | fac = 1.0 / variance(spec);
|
---|
751 | }
|
---|
752 |
|
---|
753 | // Normalize spectrum (without OTF mask) and accumulate
|
---|
754 |
|
---|
755 | const MaskedArray<Float> spec(fac*itDataVec.vector(), itMaskVec.vector());
|
---|
756 | vecSum += spec;
|
---|
757 | varSum += fac;
|
---|
758 |
|
---|
759 | // Next
|
---|
760 |
|
---|
761 | itDataVec.next();
|
---|
762 | itMaskVec.next();
|
---|
763 | }
|
---|
764 | }
|
---|
765 |
|
---|
766 | // Normalize summed spectrum
|
---|
767 |
|
---|
768 | vecSum /= varSum;
|
---|
769 |
|
---|
770 | // FInd position in input data array. We are iterating by pol-channel
|
---|
771 | // plane so all that will change is beam and IF and that's what we want.
|
---|
772 |
|
---|
773 | IPosition pos = itDataPlane.pos();
|
---|
774 |
|
---|
775 | // Write out data. This is a bit messy. We have to reform the Vector
|
---|
776 | // accumulator into an Array of shape (1,1,1,nChan)
|
---|
777 |
|
---|
778 | start = pos;
|
---|
779 | end = pos;
|
---|
780 | end(asap::ChanAxis) = nChan-1;
|
---|
781 | outData(start,end) = vecSum.getArray().reform(vecShapeOut);
|
---|
782 | outMask(start,end) = vecSum.getMask().reform(vecShapeOut);
|
---|
783 |
|
---|
784 | // Step to next beam/IF combination
|
---|
785 |
|
---|
786 | itDataPlane.next();
|
---|
787 | itMaskPlane.next();
|
---|
788 | }
|
---|
789 |
|
---|
790 | // Generate output container and write it to output table
|
---|
791 |
|
---|
792 | SDContainer sc = in.getSDContainer();
|
---|
793 | sc.resize(shapeOut);
|
---|
794 | //
|
---|
795 | putDataInSDC(sc, outData, outMask);
|
---|
796 | pTabOut->putSDContainer(sc);
|
---|
797 | }
|
---|
798 | //
|
---|
799 | return pTabOut;
|
---|
800 | }
|
---|
801 |
|
---|
802 |
|
---|
803 | SDMemTable* SDMath::smooth(const SDMemTable& in,
|
---|
804 | const casa::String& kernelType,
|
---|
805 | casa::Float width, Bool doAll) const
|
---|
806 | {
|
---|
807 |
|
---|
808 | // Number of channels
|
---|
809 |
|
---|
810 | const uInt chanAxis = asap::ChanAxis; // Spectral axis
|
---|
811 | SDHeader sh = in.getSDHeader();
|
---|
812 | const uInt nChan = sh.nchan;
|
---|
813 |
|
---|
814 | // Generate Kernel
|
---|
815 |
|
---|
816 | VectorKernel::KernelTypes type = VectorKernel::toKernelType(kernelType);
|
---|
817 | Vector<Float> kernel = VectorKernel::make(type, width, nChan, True, False);
|
---|
818 |
|
---|
819 | // Generate Convolver
|
---|
820 |
|
---|
821 | IPosition shape(1,nChan);
|
---|
822 | Convolver<Float> conv(kernel, shape);
|
---|
823 |
|
---|
824 | // New Table
|
---|
825 |
|
---|
826 | SDMemTable* pTabOut = new SDMemTable(in,True);
|
---|
827 |
|
---|
828 | // Get cursor location
|
---|
829 |
|
---|
830 | IPosition start, end;
|
---|
831 | getCursorLocation(start, end, in);
|
---|
832 | //
|
---|
833 | IPosition shapeOut(4,1);
|
---|
834 |
|
---|
835 | // Output Vectors
|
---|
836 |
|
---|
837 | Vector<Float> valuesOut(nChan);
|
---|
838 | Vector<Bool> maskOut(nChan);
|
---|
839 |
|
---|
840 | // Loop over rows in Table
|
---|
841 |
|
---|
842 | for (uInt ri=0; ri < in.nRow(); ++ri) {
|
---|
843 |
|
---|
844 | // Get copy of data
|
---|
845 |
|
---|
846 | const MaskedArray<Float>& dataIn(in.rowAsMaskedArray(ri));
|
---|
847 | AlwaysAssert(dataIn.shape()(asap::ChanAxis)==nChan, AipsError);
|
---|
848 | //
|
---|
849 | Array<Float> valuesIn = dataIn.getArray();
|
---|
850 | Array<Bool> maskIn = dataIn.getMask();
|
---|
851 |
|
---|
852 | // Branch depending on whether we smooth all locations or just
|
---|
853 | // those pointed at by the current selection cursor
|
---|
854 |
|
---|
855 | if (doAll) {
|
---|
856 | uInt axis = asap::ChanAxis;
|
---|
857 | VectorIterator<Float> itValues(valuesIn, axis);
|
---|
858 | VectorIterator<Bool> itMask(maskIn, axis);
|
---|
859 | while (!itValues.pastEnd()) {
|
---|
860 |
|
---|
861 | // Smooth
|
---|
862 | if (kernelType==VectorKernel::HANNING) {
|
---|
863 | mathutil::hanning(valuesOut, maskOut, itValues.vector(), itMask.vector());
|
---|
864 | itMask.vector() = maskOut;
|
---|
865 | } else {
|
---|
866 | mathutil::replaceMaskByZero(itValues.vector(), itMask.vector());
|
---|
867 | conv.linearConv(valuesOut, itValues.vector());
|
---|
868 | }
|
---|
869 | //
|
---|
870 | itValues.vector() = valuesOut;
|
---|
871 | //
|
---|
872 | itValues.next();
|
---|
873 | itMask.next();
|
---|
874 | }
|
---|
875 | } else {
|
---|
876 |
|
---|
877 | // Set multi-dim Vector shape
|
---|
878 |
|
---|
879 | shapeOut(asap::ChanAxis) = valuesIn.shape()(chanAxis);
|
---|
880 |
|
---|
881 | // Stuff about with shapes so that we don't have conformance run-time errors
|
---|
882 |
|
---|
883 | Vector<Float> valuesIn2 = valuesIn(start,end).nonDegenerate();
|
---|
884 | Vector<Bool> maskIn2 = maskIn(start,end).nonDegenerate();
|
---|
885 |
|
---|
886 | // Smooth
|
---|
887 |
|
---|
888 | if (kernelType==VectorKernel::HANNING) {
|
---|
889 | mathutil::hanning(valuesOut, maskOut, valuesIn2, maskIn2);
|
---|
890 | maskIn(start,end) = maskOut.reform(shapeOut);
|
---|
891 | } else {
|
---|
892 | mathutil::replaceMaskByZero(valuesIn2, maskIn2);
|
---|
893 | conv.linearConv(valuesOut, valuesIn2);
|
---|
894 | }
|
---|
895 | //
|
---|
896 | valuesIn(start,end) = valuesOut.reform(shapeOut);
|
---|
897 | }
|
---|
898 |
|
---|
899 | // Create and put back
|
---|
900 |
|
---|
901 | SDContainer sc = in.getSDContainer(ri);
|
---|
902 | putDataInSDC(sc, valuesIn, maskIn);
|
---|
903 | //
|
---|
904 | pTabOut->putSDContainer(sc);
|
---|
905 | }
|
---|
906 | //
|
---|
907 | return pTabOut;
|
---|
908 | }
|
---|
909 |
|
---|
910 |
|
---|
911 |
|
---|
912 | SDMemTable* SDMath::convertFlux (const SDMemTable& in, Float a, Float eta, Bool doAll) const
|
---|
913 | //
|
---|
914 | // As it is, this function could be implemented with 'simpleOperate'
|
---|
915 | // However, I anticipate that eventually we will look the conversion
|
---|
916 | // values up in a Table and apply them in a frequency dependent way,
|
---|
917 | // so I have implemented it fully here
|
---|
918 | //
|
---|
919 | {
|
---|
920 | SDHeader sh = in.getSDHeader();
|
---|
921 | SDMemTable* pTabOut = new SDMemTable(in, True);
|
---|
922 |
|
---|
923 | // FInd out how to convert values into Jy and K (e.g. units might be mJy or mK)
|
---|
924 | // Also automatically find out what we are converting to according to the
|
---|
925 | // flux unit
|
---|
926 |
|
---|
927 | Unit fluxUnit(sh.fluxunit);
|
---|
928 | Unit K(String("K"));
|
---|
929 | Unit JY(String("Jy"));
|
---|
930 | //
|
---|
931 | Bool toKelvin = True;
|
---|
932 | Double inFac = 1.0;
|
---|
933 | if (fluxUnit==JY) {
|
---|
934 | cerr << "Converting to K" << endl;
|
---|
935 | //
|
---|
936 | Quantum<Double> t(1.0,fluxUnit);
|
---|
937 | Quantum<Double> t2 = t.get(JY);
|
---|
938 | inFac = (t2 / t).getValue();
|
---|
939 | //
|
---|
940 | toKelvin = True;
|
---|
941 | sh.fluxunit = "K";
|
---|
942 | } else if (fluxUnit==K) {
|
---|
943 | cerr << "Converting to Jy" << endl;
|
---|
944 | //
|
---|
945 | Quantum<Double> t(1.0,fluxUnit);
|
---|
946 | Quantum<Double> t2 = t.get(K);
|
---|
947 | inFac = (t2 / t).getValue();
|
---|
948 | //
|
---|
949 | toKelvin = False;
|
---|
950 | sh.fluxunit = "Jy";
|
---|
951 | } else {
|
---|
952 | throw(AipsError("Unrecognized brightness units in Table - must be consistent with Jy or K"));
|
---|
953 | }
|
---|
954 | pTabOut->putSDHeader(sh);
|
---|
955 |
|
---|
956 | // Compute conversion factor. 'a' and 'eta' are really frequency, time and
|
---|
957 | // telescope dependent and should be looked// up in a table
|
---|
958 |
|
---|
959 | Float factor = 2.0 * inFac * 1.0e-7 * 1.0e26 *
|
---|
960 | QC::k.getValue(Unit(String("erg/K"))) / a / eta;
|
---|
961 | if (toKelvin) {
|
---|
962 | factor = 1.0 / factor;
|
---|
963 | }
|
---|
964 | cerr << "Applying conversion factor = " << factor << endl;
|
---|
965 |
|
---|
966 | // Generate correction vector. Apply same factor regardless
|
---|
967 | // of beam/pol/IF. This will need to change somewhen.
|
---|
968 |
|
---|
969 | Vector<Float> factors(in.nRow(), factor);
|
---|
970 |
|
---|
971 | // Correct
|
---|
972 |
|
---|
973 | correctFromVector (pTabOut, in, doAll, factors);
|
---|
974 | //
|
---|
975 | return pTabOut;
|
---|
976 | }
|
---|
977 |
|
---|
978 |
|
---|
979 | SDMemTable* SDMath::gainElevation (const SDMemTable& in, const Vector<Float>& coeffs,
|
---|
980 | const String& fileName,
|
---|
981 | const String& methodStr, Bool doAll) const
|
---|
982 | {
|
---|
983 |
|
---|
984 | // Get header and clone output table
|
---|
985 |
|
---|
986 | SDHeader sh = in.getSDHeader();
|
---|
987 | SDMemTable* pTabOut = new SDMemTable(in, True);
|
---|
988 |
|
---|
989 | // Get elevation data from SDMemTable and convert to degrees
|
---|
990 |
|
---|
991 | const Table& tab = in.table();
|
---|
992 | ROScalarColumn<Float> elev(tab, "ELEVATION");
|
---|
993 | Vector<Float> x = elev.getColumn();
|
---|
994 | x *= Float(180 / C::pi);
|
---|
995 | //
|
---|
996 | const uInt nC = coeffs.nelements();
|
---|
997 | if (fileName.length()>0 && nC>0) {
|
---|
998 | throw(AipsError("You must choose either polynomial coefficients or an ascii file, not both"));
|
---|
999 | }
|
---|
1000 |
|
---|
1001 | // Correct
|
---|
1002 |
|
---|
1003 | if (nC>0 || fileName.length()==0) {
|
---|
1004 |
|
---|
1005 | // Find instrument
|
---|
1006 |
|
---|
1007 | Bool throwIt = True;
|
---|
1008 | Instrument inst = SDMemTable::convertInstrument (sh.antennaname, throwIt);
|
---|
1009 |
|
---|
1010 | // Set polynomial
|
---|
1011 |
|
---|
1012 | Polynomial<Float>* pPoly = 0;
|
---|
1013 | Vector<Float> coeff;
|
---|
1014 | String msg;
|
---|
1015 | if (nC>0) {
|
---|
1016 | pPoly = new Polynomial<Float>(nC);
|
---|
1017 | coeff = coeffs;
|
---|
1018 | msg = String("user");
|
---|
1019 | } else {
|
---|
1020 | if (inst==PKSMULTIBEAM) {
|
---|
1021 | } else if (inst==PKSSINGLEBEAM) {
|
---|
1022 | } else if (inst==TIDBINBILLA) {
|
---|
1023 | pPoly = new Polynomial<Float>(3);
|
---|
1024 | coeff.resize(3);
|
---|
1025 | coeff(0) = 3.58788e-1;
|
---|
1026 | coeff(1) = 2.87243e-2;
|
---|
1027 | coeff(2) = -3.219093e-4;
|
---|
1028 | } else if (inst==MOPRA) {
|
---|
1029 | }
|
---|
1030 | msg = String("built in");
|
---|
1031 | }
|
---|
1032 | //
|
---|
1033 | if (coeff.nelements()>0) {
|
---|
1034 | pPoly->setCoefficients(coeff);
|
---|
1035 | } else {
|
---|
1036 | throw(AipsError("There is no known gain-el polynomial known for this instrument"));
|
---|
1037 | }
|
---|
1038 | //
|
---|
1039 | cerr << "Making polynomial correction with " << msg << " coefficients" << endl;
|
---|
1040 | const uInt nRow = in.nRow();
|
---|
1041 | Vector<Float> factor(nRow);
|
---|
1042 | for (uInt i=0; i<nRow; i++) {
|
---|
1043 | factor[i] = (*pPoly)(x[i]);
|
---|
1044 | }
|
---|
1045 | delete pPoly;
|
---|
1046 | //
|
---|
1047 | correctFromVector (pTabOut, in, doAll, factor);
|
---|
1048 | } else {
|
---|
1049 |
|
---|
1050 | // Indicate which columns to read from ascii file
|
---|
1051 |
|
---|
1052 | String col0("ELEVATION");
|
---|
1053 | String col1("FACTOR");
|
---|
1054 |
|
---|
1055 | // Read and correct
|
---|
1056 |
|
---|
1057 | cerr << "Making correction from ascii Table" << endl;
|
---|
1058 | correctFromAsciiTable (pTabOut, in, fileName, col0, col1,
|
---|
1059 | methodStr, doAll, x);
|
---|
1060 | }
|
---|
1061 | //
|
---|
1062 | return pTabOut;
|
---|
1063 | }
|
---|
1064 |
|
---|
1065 |
|
---|
1066 |
|
---|
1067 | SDMemTable* SDMath::opacity (const SDMemTable& in, Float tau, Bool doAll) const
|
---|
1068 | {
|
---|
1069 |
|
---|
1070 | // Get header and clone output table
|
---|
1071 |
|
---|
1072 | SDHeader sh = in.getSDHeader();
|
---|
1073 | SDMemTable* pTabOut = new SDMemTable(in, True);
|
---|
1074 |
|
---|
1075 | // Get elevation data from SDMemTable and convert to degrees
|
---|
1076 |
|
---|
1077 | const Table& tab = in.table();
|
---|
1078 | ROScalarColumn<Float> elev(tab, "ELEVATION");
|
---|
1079 | Vector<Float> zDist = elev.getColumn();
|
---|
1080 | zDist = Float(C::pi_2) - zDist;
|
---|
1081 |
|
---|
1082 | // Generate correction factor
|
---|
1083 |
|
---|
1084 | const uInt nRow = in.nRow();
|
---|
1085 | Vector<Float> factor(nRow);
|
---|
1086 | Vector<Float> factor2(nRow);
|
---|
1087 | for (uInt i=0; i<nRow; i++) {
|
---|
1088 | factor[i] = exp(tau)/cos(zDist[i]);
|
---|
1089 | }
|
---|
1090 |
|
---|
1091 | // Correct
|
---|
1092 |
|
---|
1093 | correctFromVector (pTabOut, in, doAll, factor);
|
---|
1094 | //
|
---|
1095 | return pTabOut;
|
---|
1096 | }
|
---|
1097 |
|
---|
1098 |
|
---|
1099 |
|
---|
1100 |
|
---|
1101 | // 'private' functions
|
---|
1102 |
|
---|
1103 | SDMemTable* SDMath::velocityAlign (const SDMemTable& in,
|
---|
1104 | MFrequency::Types velSystem,
|
---|
1105 | const String& velUnit,
|
---|
1106 | MDoppler::Types doppler,
|
---|
1107 | const String& refTime) const
|
---|
1108 | {
|
---|
1109 | // Get Header
|
---|
1110 |
|
---|
1111 | SDHeader sh = in.getSDHeader();
|
---|
1112 | const uInt nChan = sh.nchan;
|
---|
1113 | const uInt nRows = in.nRow();
|
---|
1114 |
|
---|
1115 | // Get Table reference
|
---|
1116 |
|
---|
1117 | const Table& tabIn = in.table();
|
---|
1118 |
|
---|
1119 | // Get Columns from Table
|
---|
1120 |
|
---|
1121 | ROScalarColumn<Double> mjdCol(tabIn, "TIME");
|
---|
1122 | ROScalarColumn<String> srcCol(tabIn, "SRCNAME");
|
---|
1123 | ROArrayColumn<uInt> fqIDCol(tabIn, "FREQID");
|
---|
1124 | //
|
---|
1125 | Vector<Double> times = mjdCol.getColumn();
|
---|
1126 | Vector<String> srcNames = srcCol.getColumn();
|
---|
1127 | Vector<uInt> freqID;
|
---|
1128 |
|
---|
1129 | // Generate Source table
|
---|
1130 |
|
---|
1131 | Vector<String> srcTab;
|
---|
1132 | Vector<uInt> srcIdx, firstRow;
|
---|
1133 | generateSourceTable (srcTab, srcIdx, firstRow, srcNames);
|
---|
1134 | const uInt nSrcTab = srcTab.nelements();
|
---|
1135 | cerr << "Found " << srcTab.nelements() << " sources to align " << endl;
|
---|
1136 |
|
---|
1137 | // Set reference Epoch to time of first row or given String
|
---|
1138 |
|
---|
1139 | Unit DAY(String("d"));
|
---|
1140 | MEpoch::Ref epochRef(in.getTimeReference());
|
---|
1141 | MEpoch refEpoch;
|
---|
1142 | if (refTime.length()>0) {
|
---|
1143 | refEpoch = epochFromString(refTime, in.getTimeReference());
|
---|
1144 | } else {
|
---|
1145 | Quantum<Double> tQ(times[0], DAY);
|
---|
1146 | MVEpoch mve(tQ);
|
---|
1147 | refEpoch = MEpoch(mve, epochRef);
|
---|
1148 | }
|
---|
1149 | cerr << "Aligning at reference Epoch " << formatEpoch(refEpoch) << endl;
|
---|
1150 |
|
---|
1151 | // Set Reference Position
|
---|
1152 |
|
---|
1153 | Vector<Double> antPos = sh.antennaposition;
|
---|
1154 | MVPosition mvpos(antPos[0], antPos[1], antPos[2]);
|
---|
1155 | MPosition refPos(mvpos);
|
---|
1156 |
|
---|
1157 | // Get Frequency Table
|
---|
1158 |
|
---|
1159 | SDFrequencyTable fTab = in.getSDFreqTable();
|
---|
1160 | const uInt nFreqIDs = fTab.length();
|
---|
1161 |
|
---|
1162 | // Create VelocityAligner Block. One VA for each possible
|
---|
1163 | // source/freqID combination
|
---|
1164 |
|
---|
1165 | PtrBlock<VelocityAligner<Float>* > vA(nFreqIDs*nSrcTab);
|
---|
1166 | for (uInt fqID=0; fqID<nFreqIDs; fqID++) {
|
---|
1167 | SpectralCoordinate sC = in.getCoordinate(fqID);
|
---|
1168 | for (uInt iSrc=0; iSrc<nSrcTab; iSrc++) {
|
---|
1169 | MDirection refDir = in.getDirection(firstRow[iSrc]);
|
---|
1170 | uInt idx = (iSrc*nFreqIDs) + fqID;
|
---|
1171 | vA[idx] = new VelocityAligner<Float>(sC, nChan, refEpoch, refDir, refPos,
|
---|
1172 | velUnit, doppler, velSystem);
|
---|
1173 | }
|
---|
1174 | }
|
---|
1175 |
|
---|
1176 | // New output Table
|
---|
1177 |
|
---|
1178 | SDMemTable* pTabOut = new SDMemTable(in,True);
|
---|
1179 |
|
---|
1180 | // Loop over rows in Table
|
---|
1181 |
|
---|
1182 | const IPosition polChanAxes(2, asap::PolAxis, asap::ChanAxis);
|
---|
1183 | VelocityAligner<Float>::Method method = VelocityAligner<Float>::LINEAR;
|
---|
1184 | Bool extrapolate=False;
|
---|
1185 | Bool useCachedAbcissa = False;
|
---|
1186 | Bool first = True;
|
---|
1187 | Bool ok;
|
---|
1188 | Vector<Float> yOut;
|
---|
1189 | Vector<Bool> maskOut;
|
---|
1190 | uInt ifIdx, vaIdx;
|
---|
1191 | //
|
---|
1192 | for (uInt iRow=0; iRow<nRows; ++iRow) {
|
---|
1193 | if (iRow%10==0) {
|
---|
1194 | cerr << "Processing row " << iRow << endl;
|
---|
1195 | }
|
---|
1196 |
|
---|
1197 | // Get EPoch
|
---|
1198 |
|
---|
1199 | Quantum<Double> tQ2(times[iRow],DAY);
|
---|
1200 | MVEpoch mv2(tQ2);
|
---|
1201 | MEpoch epoch(mv2, epochRef);
|
---|
1202 |
|
---|
1203 | // Get FreqID vector. One freqID per IF
|
---|
1204 |
|
---|
1205 | fqIDCol.get(iRow, freqID);
|
---|
1206 |
|
---|
1207 | // Get copy of data
|
---|
1208 |
|
---|
1209 | const MaskedArray<Float>& mArrIn(in.rowAsMaskedArray(iRow));
|
---|
1210 | Array<Float> values = mArrIn.getArray();
|
---|
1211 | Array<Bool> mask = mArrIn.getMask();
|
---|
1212 |
|
---|
1213 | // cerr << "values in = " << values(IPosition(4,0,0,0,0),IPosition(4,0,0,0,9)) << endl;
|
---|
1214 |
|
---|
1215 | // For each row, the Velocity abcissa will be the same regardless
|
---|
1216 | // of polarization. For all other axes (IF and BEAM) the abcissa
|
---|
1217 | // will change. So we iterate through the data by pol-chan planes
|
---|
1218 | // to mimimize the work. At this point, I think the Direction
|
---|
1219 | // is stored as the same for each beam. DOn't know where the
|
---|
1220 | // offsets are or what to do about them right now. For now
|
---|
1221 | // all beams get same position and velocoity abcissa.
|
---|
1222 |
|
---|
1223 | ArrayIterator<Float> itValuesPlane(values, polChanAxes);
|
---|
1224 | ArrayIterator<Bool> itMaskPlane(mask, polChanAxes);
|
---|
1225 | while (!itValuesPlane.pastEnd()) {
|
---|
1226 |
|
---|
1227 | // Find the IF index and then the VA PtrBlock index
|
---|
1228 |
|
---|
1229 | const IPosition& pos = itValuesPlane.pos();
|
---|
1230 | ifIdx = pos(asap::IFAxis);
|
---|
1231 | vaIdx = (srcIdx[iRow]*nFreqIDs) + freqID[ifIdx];
|
---|
1232 | //
|
---|
1233 | VectorIterator<Float> itValuesVec(itValuesPlane.array(), 1);
|
---|
1234 | VectorIterator<Bool> itMaskVec(itMaskPlane.array(), 1);
|
---|
1235 | //
|
---|
1236 | first = True;
|
---|
1237 | useCachedAbcissa=False;
|
---|
1238 | while (!itValuesVec.pastEnd()) {
|
---|
1239 | ok = vA[vaIdx]->align (yOut, maskOut, itValuesVec.vector(),
|
---|
1240 | itMaskVec.vector(), epoch, useCachedAbcissa,
|
---|
1241 | method, extrapolate);
|
---|
1242 | itValuesVec.vector() = yOut;
|
---|
1243 | itMaskVec.vector() = maskOut;
|
---|
1244 | //
|
---|
1245 | itValuesVec.next();
|
---|
1246 | itMaskVec.next();
|
---|
1247 | //
|
---|
1248 | if (first) {
|
---|
1249 | useCachedAbcissa = True;
|
---|
1250 | first = False;
|
---|
1251 | }
|
---|
1252 | }
|
---|
1253 | //
|
---|
1254 | itValuesPlane.next();
|
---|
1255 | itMaskPlane.next();
|
---|
1256 | }
|
---|
1257 |
|
---|
1258 | // cerr << "values out = " << values(IPosition(4,0,0,0,0),IPosition(4,0,0,0,9)) << endl;
|
---|
1259 |
|
---|
1260 | // Create and put back
|
---|
1261 |
|
---|
1262 | SDContainer sc = in.getSDContainer(iRow);
|
---|
1263 | putDataInSDC(sc, values, mask);
|
---|
1264 | //
|
---|
1265 | pTabOut->putSDContainer(sc);
|
---|
1266 | }
|
---|
1267 |
|
---|
1268 | // Clean up PointerBlock
|
---|
1269 |
|
---|
1270 | for (uInt i=0; i<vA.nelements(); i++) delete vA[i];
|
---|
1271 | //
|
---|
1272 | return pTabOut;
|
---|
1273 | }
|
---|
1274 |
|
---|
1275 |
|
---|
1276 | void SDMath::fillSDC(SDContainer& sc,
|
---|
1277 | const Array<Bool>& mask,
|
---|
1278 | const Array<Float>& data,
|
---|
1279 | const Array<Float>& tSys,
|
---|
1280 | Int scanID, Double timeStamp,
|
---|
1281 | Double interval, const String& sourceName,
|
---|
1282 | const Vector<uInt>& freqID) const
|
---|
1283 | {
|
---|
1284 | // Data and mask
|
---|
1285 |
|
---|
1286 | putDataInSDC(sc, data, mask);
|
---|
1287 |
|
---|
1288 | // TSys
|
---|
1289 |
|
---|
1290 | sc.putTsys(tSys);
|
---|
1291 |
|
---|
1292 | // Time things
|
---|
1293 |
|
---|
1294 | sc.timestamp = timeStamp;
|
---|
1295 | sc.interval = interval;
|
---|
1296 | sc.scanid = scanID;
|
---|
1297 | //
|
---|
1298 | sc.sourcename = sourceName;
|
---|
1299 | sc.putFreqMap(freqID);
|
---|
1300 | }
|
---|
1301 |
|
---|
1302 | void SDMath::normalize(MaskedArray<Float>& sum,
|
---|
1303 | const Array<Float>& sumSq,
|
---|
1304 | const Array<Float>& nPts,
|
---|
1305 | WeightType wtType, Int axis,
|
---|
1306 | Int nAxesSub) const
|
---|
1307 | {
|
---|
1308 | IPosition pos2(nAxesSub,0);
|
---|
1309 | //
|
---|
1310 | if (wtType==NONE) {
|
---|
1311 |
|
---|
1312 | // We just average by the number of points accumulated.
|
---|
1313 | // We need to make a MA out of nPts so that no divide by
|
---|
1314 | // zeros occur
|
---|
1315 |
|
---|
1316 | MaskedArray<Float> t(nPts, (nPts>Float(0.0)));
|
---|
1317 | sum /= t;
|
---|
1318 | } else if (wtType==VAR) {
|
---|
1319 |
|
---|
1320 | // Normalize each spectrum by sum(1/var) where the variance
|
---|
1321 | // is worked out for each spectrum
|
---|
1322 |
|
---|
1323 | Array<Float>& data = sum.getRWArray();
|
---|
1324 | VectorIterator<Float> itData(data, axis);
|
---|
1325 | while (!itData.pastEnd()) {
|
---|
1326 | pos2 = itData.pos().getFirst(nAxesSub);
|
---|
1327 | itData.vector() /= sumSq(pos2);
|
---|
1328 | itData.next();
|
---|
1329 | }
|
---|
1330 | } else if (wtType==TSYS) {
|
---|
1331 | }
|
---|
1332 | }
|
---|
1333 |
|
---|
1334 |
|
---|
1335 | void SDMath::accumulate(Double& timeSum, Double& intSum, Int& nAccum,
|
---|
1336 | MaskedArray<Float>& sum, Array<Float>& sumSq,
|
---|
1337 | Array<Float>& nPts, Array<Float>& tSysSum,
|
---|
1338 | const Array<Float>& tSys, const Array<Float>& nInc,
|
---|
1339 | const Vector<Bool>& mask, Double time, Double interval,
|
---|
1340 | const Block<CountedPtr<SDMemTable> >& in,
|
---|
1341 | uInt iTab, uInt iRow, uInt axis,
|
---|
1342 | uInt nAxesSub, Bool useMask,
|
---|
1343 | WeightType wtType) const
|
---|
1344 | {
|
---|
1345 |
|
---|
1346 | // Get data
|
---|
1347 |
|
---|
1348 | MaskedArray<Float> dataIn(in[iTab]->rowAsMaskedArray(iRow));
|
---|
1349 | Array<Float>& valuesIn = dataIn.getRWArray(); // writable reference
|
---|
1350 | const Array<Bool>& maskIn = dataIn.getMask(); // RO reference
|
---|
1351 | //
|
---|
1352 | if (wtType==NONE) {
|
---|
1353 | const MaskedArray<Float> n(nInc,dataIn.getMask());
|
---|
1354 | nPts += n; // Only accumulates where mask==T
|
---|
1355 | } else if (wtType==VAR) {
|
---|
1356 |
|
---|
1357 | // We are going to average the data, weighted by the noise for each pol, beam and IF.
|
---|
1358 | // So therefore we need to iterate through by spectrum (axis 3)
|
---|
1359 |
|
---|
1360 | VectorIterator<Float> itData(valuesIn, axis);
|
---|
1361 | ReadOnlyVectorIterator<Bool> itMask(maskIn, axis);
|
---|
1362 | Float fac = 1.0;
|
---|
1363 | IPosition pos(nAxesSub,0);
|
---|
1364 | //
|
---|
1365 | while (!itData.pastEnd()) {
|
---|
1366 |
|
---|
1367 | // Make MaskedArray of Vector, optionally apply OTF mask, and find scaling factor
|
---|
1368 |
|
---|
1369 | if (useMask) {
|
---|
1370 | MaskedArray<Float> tmp(itData.vector(),mask&&itMask.vector());
|
---|
1371 | fac = 1.0/variance(tmp);
|
---|
1372 | } else {
|
---|
1373 | MaskedArray<Float> tmp(itData.vector(),itMask.vector());
|
---|
1374 | fac = 1.0/variance(tmp);
|
---|
1375 | }
|
---|
1376 |
|
---|
1377 | // Scale data
|
---|
1378 |
|
---|
1379 | itData.vector() *= fac; // Writes back into 'dataIn'
|
---|
1380 | //
|
---|
1381 | // Accumulate variance per if/pol/beam averaged over spectrum
|
---|
1382 | // This method to get pos2 from itData.pos() is only valid
|
---|
1383 | // because the spectral axis is the last one (so we can just
|
---|
1384 | // copy the first nAXesSub positions out)
|
---|
1385 |
|
---|
1386 | pos = itData.pos().getFirst(nAxesSub);
|
---|
1387 | sumSq(pos) += fac;
|
---|
1388 | //
|
---|
1389 | itData.next();
|
---|
1390 | itMask.next();
|
---|
1391 | }
|
---|
1392 | } else if (wtType==TSYS) {
|
---|
1393 | }
|
---|
1394 |
|
---|
1395 | // Accumulate sum of (possibly scaled) data
|
---|
1396 |
|
---|
1397 | sum += dataIn;
|
---|
1398 |
|
---|
1399 | // Accumulate Tsys, time, and interval
|
---|
1400 |
|
---|
1401 | tSysSum += tSys;
|
---|
1402 | timeSum += time;
|
---|
1403 | intSum += interval;
|
---|
1404 | nAccum += 1;
|
---|
1405 | }
|
---|
1406 |
|
---|
1407 |
|
---|
1408 |
|
---|
1409 |
|
---|
1410 | void SDMath::getCursorLocation(IPosition& start, IPosition& end,
|
---|
1411 | const SDMemTable& in) const
|
---|
1412 | {
|
---|
1413 | const uInt nDim = 4;
|
---|
1414 | const uInt i = in.getBeam();
|
---|
1415 | const uInt j = in.getIF();
|
---|
1416 | const uInt k = in.getPol();
|
---|
1417 | const uInt n = in.nChan();
|
---|
1418 | //
|
---|
1419 | start.resize(nDim);
|
---|
1420 | start(0) = i;
|
---|
1421 | start(1) = j;
|
---|
1422 | start(2) = k;
|
---|
1423 | start(3) = 0;
|
---|
1424 | //
|
---|
1425 | end.resize(nDim);
|
---|
1426 | end(0) = i;
|
---|
1427 | end(1) = j;
|
---|
1428 | end(2) = k;
|
---|
1429 | end(3) = n-1;
|
---|
1430 | }
|
---|
1431 |
|
---|
1432 |
|
---|
1433 | void SDMath::convertWeightString(WeightType& wtType, const String& weightStr) const
|
---|
1434 | {
|
---|
1435 | String tStr(weightStr);
|
---|
1436 | tStr.upcase();
|
---|
1437 | if (tStr.contains(String("NONE"))) {
|
---|
1438 | wtType = NONE;
|
---|
1439 | } else if (tStr.contains(String("VAR"))) {
|
---|
1440 | wtType = VAR;
|
---|
1441 | } else if (tStr.contains(String("TSYS"))) {
|
---|
1442 | wtType = TSYS;
|
---|
1443 | throw(AipsError("T_sys weighting not yet implemented"));
|
---|
1444 | } else {
|
---|
1445 | throw(AipsError("Unrecognized weighting type"));
|
---|
1446 | }
|
---|
1447 | }
|
---|
1448 |
|
---|
1449 | void SDMath::convertInterpString(Int& type, const String& interp) const
|
---|
1450 | {
|
---|
1451 | String tStr(interp);
|
---|
1452 | tStr.upcase();
|
---|
1453 | if (tStr.contains(String("NEAR"))) {
|
---|
1454 | type = InterpolateArray1D<Float,Float>::nearestNeighbour;
|
---|
1455 | } else if (tStr.contains(String("LIN"))) {
|
---|
1456 | type = InterpolateArray1D<Float,Float>::linear;
|
---|
1457 | } else if (tStr.contains(String("CUB"))) {
|
---|
1458 | type = InterpolateArray1D<Float,Float>::cubic;
|
---|
1459 | } else if (tStr.contains(String("SPL"))) {
|
---|
1460 | type = InterpolateArray1D<Float,Float>::spline;
|
---|
1461 | } else {
|
---|
1462 | throw(AipsError("Unrecognized interpolation type"));
|
---|
1463 | }
|
---|
1464 | }
|
---|
1465 |
|
---|
1466 | void SDMath::putDataInSDC(SDContainer& sc, const Array<Float>& data,
|
---|
1467 | const Array<Bool>& mask) const
|
---|
1468 | {
|
---|
1469 | sc.putSpectrum(data);
|
---|
1470 | //
|
---|
1471 | Array<uChar> outflags(data.shape());
|
---|
1472 | convertArray(outflags,!mask);
|
---|
1473 | sc.putFlags(outflags);
|
---|
1474 | }
|
---|
1475 |
|
---|
1476 | Table SDMath::readAsciiFile (const String& fileName) const
|
---|
1477 | {
|
---|
1478 | String formatString;
|
---|
1479 | Table tbl = readAsciiTable (formatString, Table::Memory, fileName, "", "", False);
|
---|
1480 | return tbl;
|
---|
1481 | }
|
---|
1482 |
|
---|
1483 |
|
---|
1484 |
|
---|
1485 | void SDMath::correctFromAsciiTable(SDMemTable* pTabOut,
|
---|
1486 | const SDMemTable& in, const String& fileName,
|
---|
1487 | const String& col0, const String& col1,
|
---|
1488 | const String& methodStr, Bool doAll,
|
---|
1489 | const Vector<Float>& xOut) const
|
---|
1490 | {
|
---|
1491 |
|
---|
1492 | // Read gain-elevation ascii file data into a Table.
|
---|
1493 |
|
---|
1494 | Table geTable = readAsciiFile (fileName);
|
---|
1495 | //
|
---|
1496 | correctFromTable (pTabOut, in, geTable, col0, col1, methodStr, doAll, xOut);
|
---|
1497 | }
|
---|
1498 |
|
---|
1499 | void SDMath::correctFromTable(SDMemTable* pTabOut, const SDMemTable& in,
|
---|
1500 | const Table& tTable, const String& col0,
|
---|
1501 | const String& col1,
|
---|
1502 | const String& methodStr, Bool doAll,
|
---|
1503 | const Vector<Float>& xOut) const
|
---|
1504 | {
|
---|
1505 |
|
---|
1506 | // Get data from Table
|
---|
1507 |
|
---|
1508 | ROScalarColumn<Float> geElCol(tTable, col0);
|
---|
1509 | ROScalarColumn<Float> geFacCol(tTable, col1);
|
---|
1510 | Vector<Float> xIn = geElCol.getColumn();
|
---|
1511 | Vector<Float> yIn = geFacCol.getColumn();
|
---|
1512 | Vector<Bool> maskIn(xIn.nelements(),True);
|
---|
1513 |
|
---|
1514 | // Interpolate (and extrapolate) with desired method
|
---|
1515 |
|
---|
1516 | Int method = 0;
|
---|
1517 | convertInterpString(method, methodStr);
|
---|
1518 | //
|
---|
1519 | Vector<Float> yOut;
|
---|
1520 | Vector<Bool> maskOut;
|
---|
1521 | InterpolateArray1D<Float,Float>::interpolate(yOut, maskOut, xOut,
|
---|
1522 | xIn, yIn, maskIn, method,
|
---|
1523 | True, True);
|
---|
1524 | // Apply
|
---|
1525 |
|
---|
1526 | correctFromVector (pTabOut, in, doAll, yOut);
|
---|
1527 | }
|
---|
1528 |
|
---|
1529 |
|
---|
1530 | void SDMath::correctFromVector (SDMemTable* pTabOut, const SDMemTable& in,
|
---|
1531 | Bool doAll, const Vector<Float>& factor) const
|
---|
1532 | {
|
---|
1533 |
|
---|
1534 | // For operations only on specified cursor location
|
---|
1535 |
|
---|
1536 | IPosition start, end;
|
---|
1537 | getCursorLocation(start, end, in);
|
---|
1538 |
|
---|
1539 | // Loop over rows and apply correction factor
|
---|
1540 |
|
---|
1541 | const uInt axis = asap::ChanAxis;
|
---|
1542 | for (uInt i=0; i < in.nRow(); ++i) {
|
---|
1543 |
|
---|
1544 | // Get data
|
---|
1545 |
|
---|
1546 | MaskedArray<Float> dataIn(in.rowAsMaskedArray(i));
|
---|
1547 |
|
---|
1548 | // Apply factor
|
---|
1549 |
|
---|
1550 | if (doAll) {
|
---|
1551 | dataIn *= factor[i];
|
---|
1552 | } else {
|
---|
1553 | MaskedArray<Float> dataIn2 = dataIn(start,end); // reference
|
---|
1554 | dataIn2 *= factor[i];
|
---|
1555 | }
|
---|
1556 |
|
---|
1557 | // Write out
|
---|
1558 |
|
---|
1559 | SDContainer sc = in.getSDContainer(i);
|
---|
1560 | putDataInSDC(sc, dataIn.getArray(), dataIn.getMask());
|
---|
1561 | //
|
---|
1562 | pTabOut->putSDContainer(sc);
|
---|
1563 | }
|
---|
1564 | }
|
---|
1565 |
|
---|
1566 |
|
---|
1567 | void SDMath::generateSourceTable (Vector<String>& srcTab,
|
---|
1568 | Vector<uInt>& srcIdx,
|
---|
1569 | Vector<uInt>& firstRow,
|
---|
1570 | const Vector<String>& srcNames) const
|
---|
1571 | //
|
---|
1572 | // This algorithm assumes that if there are multiple beams
|
---|
1573 | // that the source names are diffent. Oterwise we would need
|
---|
1574 | // to look atthe direction for each beam...
|
---|
1575 | //
|
---|
1576 | {
|
---|
1577 | const uInt nRow = srcNames.nelements();
|
---|
1578 | srcTab.resize(0);
|
---|
1579 | srcIdx.resize(nRow);
|
---|
1580 | firstRow.resize(0);
|
---|
1581 | //
|
---|
1582 | uInt nSrc = 0;
|
---|
1583 | for (uInt i=0; i<nRow; i++) {
|
---|
1584 | String srcName = srcNames[i];
|
---|
1585 |
|
---|
1586 | // Do we have this source already ?
|
---|
1587 |
|
---|
1588 | Int idx = -1;
|
---|
1589 | if (nSrc>0) {
|
---|
1590 | for (uInt j=0; j<nSrc; j++) {
|
---|
1591 | if (srcName==srcTab[j]) {
|
---|
1592 | idx = j;
|
---|
1593 | break;
|
---|
1594 | }
|
---|
1595 | }
|
---|
1596 | }
|
---|
1597 |
|
---|
1598 | // Add new entry if not found
|
---|
1599 |
|
---|
1600 | if (idx==-1) {
|
---|
1601 | nSrc++;
|
---|
1602 | srcTab.resize(nSrc,True);
|
---|
1603 | srcTab(nSrc-1) = srcName;
|
---|
1604 | idx = nSrc-1;
|
---|
1605 | //
|
---|
1606 | firstRow.resize(nSrc,True);
|
---|
1607 | firstRow(nSrc-1) = i; // First row for which this source occurs
|
---|
1608 | }
|
---|
1609 |
|
---|
1610 | // Set index for this row
|
---|
1611 |
|
---|
1612 | srcIdx[i] = idx;
|
---|
1613 | }
|
---|
1614 | }
|
---|
1615 |
|
---|
1616 | MEpoch SDMath::epochFromString (const String& str, MEpoch::Types timeRef) const
|
---|
1617 | {
|
---|
1618 | Quantum<Double> qt;
|
---|
1619 | if (MVTime::read(qt,str)) {
|
---|
1620 | MVEpoch mv(qt);
|
---|
1621 | MEpoch me(mv, timeRef);
|
---|
1622 | return me;
|
---|
1623 | } else {
|
---|
1624 | throw(AipsError("Invalid format for Epoch string"));
|
---|
1625 | }
|
---|
1626 | }
|
---|
1627 |
|
---|
1628 |
|
---|
1629 | String SDMath::formatEpoch(const MEpoch& epoch) const
|
---|
1630 | {
|
---|
1631 | MVTime mvt(epoch.getValue());
|
---|
1632 | return mvt.string(MVTime::YMD) + String(" (") + epoch.getRefString() + String(")");
|
---|
1633 | }
|
---|
1634 |
|
---|