[2] | 1 | //#---------------------------------------------------------------------------
|
---|
| 2 | //# SDMath.cc: A collection of single dish mathematical operations
|
---|
| 3 | //#---------------------------------------------------------------------------
|
---|
| 4 | //# Copyright (C) 2004
|
---|
[125] | 5 | //# ATNF
|
---|
[2] | 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 | //#---------------------------------------------------------------------------
|
---|
[38] | 31 | #include <vector>
|
---|
| 32 |
|
---|
[81] | 33 | #include <casa/aips.h>
|
---|
| 34 | #include <casa/BasicSL/String.h>
|
---|
| 35 | #include <casa/Arrays/IPosition.h>
|
---|
| 36 | #include <casa/Arrays/Array.h>
|
---|
[130] | 37 | #include <casa/Arrays/ArrayIter.h>
|
---|
| 38 | #include <casa/Arrays/VectorIter.h>
|
---|
[81] | 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>
|
---|
[221] | 44 | #include <casa/Containers/Block.h>
|
---|
| 45 | #include <casa/Quanta/QC.h>
|
---|
[177] | 46 | #include <casa/Utilities/Assert.h>
|
---|
[130] | 47 | #include <casa/Exceptions.h>
|
---|
[2] | 48 |
|
---|
[177] | 49 | #include <scimath/Mathematics/VectorKernel.h>
|
---|
| 50 | #include <scimath/Mathematics/Convolver.h>
|
---|
[227] | 51 | #include <scimath/Mathematics/InterpolateArray1D.h>
|
---|
[177] | 52 |
|
---|
[81] | 53 | #include <tables/Tables/Table.h>
|
---|
| 54 | #include <tables/Tables/ScalarColumn.h>
|
---|
| 55 | #include <tables/Tables/ArrayColumn.h>
|
---|
[227] | 56 | #include <tables/Tables/ReadAsciiTable.h>
|
---|
[2] | 57 |
|
---|
[130] | 58 | #include <lattices/Lattices/LatticeUtilities.h>
|
---|
| 59 | #include <lattices/Lattices/RebinLattice.h>
|
---|
[81] | 60 | #include <coordinates/Coordinates/SpectralCoordinate.h>
|
---|
[130] | 61 | #include <coordinates/Coordinates/CoordinateSystem.h>
|
---|
| 62 | #include <coordinates/Coordinates/CoordinateUtil.h>
|
---|
[221] | 63 | #include <coordinates/Coordinates/VelocityAligner.h>
|
---|
[38] | 64 |
|
---|
| 65 | #include "MathUtils.h"
|
---|
[209] | 66 | #include "Definitions.h"
|
---|
[2] | 67 | #include "SDContainer.h"
|
---|
| 68 | #include "SDMemTable.h"
|
---|
| 69 |
|
---|
| 70 | #include "SDMath.h"
|
---|
| 71 |
|
---|
[125] | 72 | using namespace casa;
|
---|
[83] | 73 | using namespace asap;
|
---|
[2] | 74 |
|
---|
[170] | 75 |
|
---|
| 76 | SDMath::SDMath()
|
---|
| 77 | {;}
|
---|
| 78 |
|
---|
[185] | 79 | SDMath::SDMath(const SDMath& other)
|
---|
[170] | 80 | {
|
---|
| 81 |
|
---|
| 82 | // No state
|
---|
| 83 |
|
---|
| 84 | }
|
---|
| 85 |
|
---|
| 86 | SDMath& SDMath::operator=(const SDMath& other)
|
---|
| 87 | {
|
---|
| 88 | if (this != &other) {
|
---|
| 89 | // No state
|
---|
| 90 | }
|
---|
| 91 | return *this;
|
---|
| 92 | }
|
---|
| 93 |
|
---|
[183] | 94 | SDMath::~SDMath()
|
---|
| 95 | {;}
|
---|
[170] | 96 |
|
---|
[183] | 97 |
|
---|
[185] | 98 | CountedPtr<SDMemTable> SDMath::average(const Block<CountedPtr<SDMemTable> >& in,
|
---|
| 99 | const Vector<Bool>& mask, Bool scanAv,
|
---|
[227] | 100 | const String& weightStr)
|
---|
[221] | 101 | //Bool alignVelocity)
|
---|
[130] | 102 | //
|
---|
[144] | 103 | // Weighted averaging of spectra from one or more Tables.
|
---|
[130] | 104 | //
|
---|
| 105 | {
|
---|
[221] | 106 | Bool alignVelocity = False;
|
---|
[2] | 107 |
|
---|
[163] | 108 | // Convert weight type
|
---|
| 109 |
|
---|
| 110 | WeightType wtType = NONE;
|
---|
[185] | 111 | convertWeightString(wtType, weightStr);
|
---|
[163] | 112 |
|
---|
[144] | 113 | // Create output Table by cloning from the first table
|
---|
[2] | 114 |
|
---|
[144] | 115 | SDMemTable* pTabOut = new SDMemTable(*in[0],True);
|
---|
[130] | 116 |
|
---|
[144] | 117 | // Setup
|
---|
[130] | 118 |
|
---|
[144] | 119 | IPosition shp = in[0]->rowAsMaskedArray(0).shape(); // Must not change
|
---|
| 120 | Array<Float> arr(shp);
|
---|
| 121 | Array<Bool> barr(shp);
|
---|
[221] | 122 | const Bool useMask = (mask.nelements() == shp(asap::ChanAxis));
|
---|
[130] | 123 |
|
---|
[144] | 124 | // Columns from Tables
|
---|
[130] | 125 |
|
---|
[144] | 126 | ROArrayColumn<Float> tSysCol;
|
---|
| 127 | ROScalarColumn<Double> mjdCol;
|
---|
| 128 | ROScalarColumn<String> srcNameCol;
|
---|
| 129 | ROScalarColumn<Double> intCol;
|
---|
| 130 | ROArrayColumn<uInt> fqIDCol;
|
---|
[130] | 131 |
|
---|
[144] | 132 | // Create accumulation MaskedArray. We accumulate for each channel,if,pol,beam
|
---|
| 133 | // Note that the mask of the accumulation array will ALWAYS remain ALL True.
|
---|
| 134 | // The MA is only used so that when data which is masked Bad is added to it,
|
---|
| 135 | // that data does not contribute.
|
---|
| 136 |
|
---|
| 137 | Array<Float> zero(shp);
|
---|
| 138 | zero=0.0;
|
---|
| 139 | Array<Bool> good(shp);
|
---|
| 140 | good = True;
|
---|
| 141 | MaskedArray<Float> sum(zero,good);
|
---|
| 142 |
|
---|
| 143 | // Counter arrays
|
---|
| 144 |
|
---|
| 145 | Array<Float> nPts(shp); // Number of points
|
---|
| 146 | nPts = 0.0;
|
---|
| 147 | Array<Float> nInc(shp); // Increment
|
---|
| 148 | nInc = 1.0;
|
---|
| 149 |
|
---|
| 150 | // Create accumulation Array for variance. We accumulate for
|
---|
| 151 | // each if,pol,beam, but average over channel. So we need
|
---|
| 152 | // a shape with one less axis dropping channels.
|
---|
| 153 |
|
---|
| 154 | const uInt nAxesSub = shp.nelements() - 1;
|
---|
| 155 | IPosition shp2(nAxesSub);
|
---|
| 156 | for (uInt i=0,j=0; i<(nAxesSub+1); i++) {
|
---|
[221] | 157 | if (i!=asap::ChanAxis) {
|
---|
[144] | 158 | shp2(j) = shp(i);
|
---|
| 159 | j++;
|
---|
| 160 | }
|
---|
[2] | 161 | }
|
---|
[144] | 162 | Array<Float> sumSq(shp2);
|
---|
| 163 | sumSq = 0.0;
|
---|
| 164 | IPosition pos2(nAxesSub,0); // For indexing
|
---|
[130] | 165 |
|
---|
[144] | 166 | // Time-related accumulators
|
---|
[130] | 167 |
|
---|
[144] | 168 | Double time;
|
---|
| 169 | Double timeSum = 0.0;
|
---|
| 170 | Double intSum = 0.0;
|
---|
| 171 | Double interval = 0.0;
|
---|
[130] | 172 |
|
---|
[144] | 173 | // To get the right shape for the Tsys accumulator we need to
|
---|
| 174 | // access a column from the first table. The shape of this
|
---|
| 175 | // array must not change
|
---|
[130] | 176 |
|
---|
[144] | 177 | Array<Float> tSysSum;
|
---|
| 178 | {
|
---|
| 179 | const Table& tabIn = in[0]->table();
|
---|
| 180 | tSysCol.attach(tabIn,"TSYS");
|
---|
| 181 | tSysSum.resize(tSysCol.shape(0));
|
---|
| 182 | }
|
---|
| 183 | tSysSum =0.0;
|
---|
| 184 | Array<Float> tSys;
|
---|
| 185 |
|
---|
| 186 | // Scan and row tracking
|
---|
| 187 |
|
---|
| 188 | Int oldScanID = 0;
|
---|
| 189 | Int outScanID = 0;
|
---|
| 190 | Int scanID = 0;
|
---|
| 191 | Int rowStart = 0;
|
---|
| 192 | Int nAccum = 0;
|
---|
| 193 | Int tableStart = 0;
|
---|
| 194 |
|
---|
| 195 | // Source and FreqID
|
---|
| 196 |
|
---|
| 197 | String sourceName, oldSourceName, sourceNameStart;
|
---|
| 198 | Vector<uInt> freqID, freqIDStart, oldFreqID;
|
---|
| 199 |
|
---|
[221] | 200 | // Velocity Aligner. We need an aligner for each Direction and FreqID
|
---|
| 201 | // combination. I don't think there is anyway to know how many
|
---|
| 202 | // directions there are.
|
---|
| 203 | // For now, assume all Tables have the same Frequency Table
|
---|
| 204 |
|
---|
| 205 | /*
|
---|
| 206 | {
|
---|
| 207 | MEpoch::Ref timeRef(MEpoch::UTC); // Should be in header
|
---|
| 208 | MDirection::Types dirRef(MDirection::J2000); // Should be in header
|
---|
| 209 | //
|
---|
| 210 | SDHeader sh = in[0].getSDHeader();
|
---|
| 211 | const uInt nChan = sh.nchan;
|
---|
| 212 | //
|
---|
| 213 | const SDFrequencyTable freqTab = in[0]->getSDFreqTable();
|
---|
| 214 | const uInt nFreqID = freqTab.length();
|
---|
| 215 | PtrBlock<const VelocityAligner<Float>* > vA(nFreqID);
|
---|
| 216 |
|
---|
| 217 | // Get first time from first table
|
---|
| 218 |
|
---|
| 219 | const Table& tabIn0 = in[0]->table();
|
---|
| 220 | mjdCol.attach(tabIn0, "TIME");
|
---|
| 221 | Double dTmp;
|
---|
| 222 | mjdCol.get(0, dTmp);
|
---|
| 223 | MVEpoch tmp2(Quantum<Double>(dTmp, Unit(String("d"))));
|
---|
| 224 | MEpoch epoch(tmp2, timeRef);
|
---|
| 225 | //
|
---|
| 226 | for (uInt freqID=0; freqID<nFreqID; freqID++) {
|
---|
| 227 | SpectralCoordinate sC = in[0]->getCoordinate(freqID);
|
---|
| 228 | vA[freqID] = new VelocityAligner<Float>(sC, nChan, epoch, const MDirection& dir,
|
---|
| 229 | const MPosition& pos, const String& velUnit,
|
---|
| 230 | MDoppler::Types velType, MFrequency::Types velFreqSystem)
|
---|
| 231 | }
|
---|
| 232 | }
|
---|
| 233 | */
|
---|
| 234 |
|
---|
[144] | 235 | // Loop over tables
|
---|
| 236 |
|
---|
| 237 | Float fac = 1.0;
|
---|
| 238 | const uInt nTables = in.nelements();
|
---|
| 239 | for (uInt iTab=0; iTab<nTables; iTab++) {
|
---|
| 240 |
|
---|
[221] | 241 | // Should check that the frequency tables don't change if doing VelocityAlignment
|
---|
| 242 |
|
---|
[144] | 243 | // Attach columns to Table
|
---|
| 244 |
|
---|
| 245 | const Table& tabIn = in[iTab]->table();
|
---|
| 246 | tSysCol.attach(tabIn, "TSYS");
|
---|
| 247 | mjdCol.attach(tabIn, "TIME");
|
---|
| 248 | srcNameCol.attach(tabIn, "SRCNAME");
|
---|
| 249 | intCol.attach(tabIn, "INTERVAL");
|
---|
| 250 | fqIDCol.attach(tabIn, "FREQID");
|
---|
| 251 |
|
---|
| 252 | // Loop over rows in Table
|
---|
| 253 |
|
---|
| 254 | const uInt nRows = in[iTab]->nRow();
|
---|
| 255 | for (uInt iRow=0; iRow<nRows; iRow++) {
|
---|
| 256 |
|
---|
| 257 | // Check conformance
|
---|
| 258 |
|
---|
| 259 | IPosition shp2 = in[iTab]->rowAsMaskedArray(iRow).shape();
|
---|
| 260 | if (!shp.isEqual(shp2)) {
|
---|
| 261 | throw (AipsError("Shapes for all rows must be the same"));
|
---|
| 262 | }
|
---|
| 263 |
|
---|
| 264 | // If we are not doing scan averages, make checks for source and
|
---|
| 265 | // frequency setup and warn if averaging across them
|
---|
| 266 |
|
---|
| 267 | // Get copy of Scan Container for this row
|
---|
| 268 |
|
---|
| 269 | SDContainer sc = in[iTab]->getSDContainer(iRow);
|
---|
| 270 | scanID = sc.scanid;
|
---|
| 271 |
|
---|
| 272 | // Get quantities from columns
|
---|
| 273 |
|
---|
| 274 | srcNameCol.getScalar(iRow, sourceName);
|
---|
| 275 | mjdCol.get(iRow, time);
|
---|
| 276 | tSysCol.get(iRow, tSys);
|
---|
| 277 | intCol.get(iRow, interval);
|
---|
| 278 | fqIDCol.get(iRow, freqID);
|
---|
| 279 |
|
---|
| 280 | // Initialize first source and freqID
|
---|
| 281 |
|
---|
| 282 | if (iRow==0 && iTab==0) {
|
---|
| 283 | sourceNameStart = sourceName;
|
---|
| 284 | freqIDStart = freqID;
|
---|
| 285 | }
|
---|
| 286 |
|
---|
| 287 | // If we are doing scan averages, see if we are at the end of an
|
---|
| 288 | // accumulation period (scan). We must check soutce names too,
|
---|
| 289 | // since we might have two tables with one scan each but different
|
---|
| 290 | // source names; we shouldn't average different sources together
|
---|
| 291 |
|
---|
| 292 | if (scanAv && ( (scanID != oldScanID) ||
|
---|
| 293 | (iRow==0 && iTab>0 && sourceName!=oldSourceName))) {
|
---|
| 294 |
|
---|
| 295 | // Normalize data in 'sum' accumulation array according to weighting scheme
|
---|
| 296 |
|
---|
[221] | 297 | normalize(sum, sumSq, nPts, wtType, asap::ChanAxis, nAxesSub);
|
---|
[144] | 298 |
|
---|
| 299 | // Fill scan container. The source and freqID come from the
|
---|
| 300 | // first row of the first table that went into this average (
|
---|
| 301 | // should be the same for all rows in the scan average)
|
---|
| 302 |
|
---|
| 303 | Float nR(nAccum);
|
---|
[185] | 304 | fillSDC(sc, sum.getMask(), sum.getArray(), tSysSum/nR, outScanID,
|
---|
[144] | 305 | timeSum/nR, intSum, sourceNameStart, freqIDStart);
|
---|
| 306 |
|
---|
| 307 | // Write container out to Table
|
---|
| 308 |
|
---|
| 309 | pTabOut->putSDContainer(sc);
|
---|
| 310 |
|
---|
| 311 | // Reset accumulators
|
---|
| 312 |
|
---|
| 313 | sum = 0.0;
|
---|
| 314 | sumSq = 0.0;
|
---|
| 315 | nAccum = 0;
|
---|
| 316 | //
|
---|
| 317 | tSysSum =0.0;
|
---|
| 318 | timeSum = 0.0;
|
---|
| 319 | intSum = 0.0;
|
---|
[221] | 320 | nPts = 0.0;
|
---|
[144] | 321 |
|
---|
| 322 | // Increment
|
---|
| 323 |
|
---|
| 324 | rowStart = iRow; // First row for next accumulation
|
---|
| 325 | tableStart = iTab; // First table for next accumulation
|
---|
| 326 | sourceNameStart = sourceName; // First source name for next accumulation
|
---|
| 327 | freqIDStart = freqID; // First FreqID for next accumulation
|
---|
| 328 | //
|
---|
| 329 | oldScanID = scanID;
|
---|
| 330 | outScanID += 1; // Scan ID for next accumulation period
|
---|
[227] | 331 | }
|
---|
[144] | 332 |
|
---|
[146] | 333 | // Accumulate
|
---|
[144] | 334 |
|
---|
[185] | 335 | accumulate(timeSum, intSum, nAccum, sum, sumSq, nPts, tSysSum,
|
---|
[221] | 336 | tSys, nInc, mask, time, interval, in, iTab, iRow, asap::ChanAxis,
|
---|
[146] | 337 | nAxesSub, useMask, wtType);
|
---|
[144] | 338 | //
|
---|
| 339 | oldSourceName = sourceName;
|
---|
| 340 | oldFreqID = freqID;
|
---|
[184] | 341 | }
|
---|
[144] | 342 | }
|
---|
| 343 |
|
---|
| 344 | // OK at this point we have accumulation data which is either
|
---|
| 345 | // - accumulated from all tables into one row
|
---|
| 346 | // or
|
---|
| 347 | // - accumulated from the last scan average
|
---|
| 348 | //
|
---|
| 349 | // Normalize data in 'sum' accumulation array according to weighting scheme
|
---|
[221] | 350 | normalize(sum, sumSq, nPts, wtType, asap::ChanAxis, nAxesSub);
|
---|
[144] | 351 |
|
---|
| 352 | // Create and fill container. The container we clone will be from
|
---|
| 353 | // the last Table and the first row that went into the current
|
---|
| 354 | // accumulation. It probably doesn't matter that much really...
|
---|
| 355 |
|
---|
| 356 | Float nR(nAccum);
|
---|
| 357 | SDContainer sc = in[tableStart]->getSDContainer(rowStart);
|
---|
[185] | 358 | fillSDC(sc, sum.getMask(), sum.getArray(), tSysSum/nR, outScanID,
|
---|
[144] | 359 | timeSum/nR, intSum, sourceNameStart, freqIDStart);
|
---|
[221] | 360 | pTabOut->putSDContainer(sc);
|
---|
[144] | 361 | //
|
---|
| 362 | return CountedPtr<SDMemTable>(pTabOut);
|
---|
[2] | 363 | }
|
---|
[9] | 364 |
|
---|
[144] | 365 |
|
---|
| 366 |
|
---|
[85] | 367 | CountedPtr<SDMemTable>
|
---|
| 368 | SDMath::quotient(const CountedPtr<SDMemTable>& on,
|
---|
[185] | 369 | const CountedPtr<SDMemTable>& off)
|
---|
| 370 | {
|
---|
[130] | 371 | //
|
---|
| 372 | // Compute quotient spectrum
|
---|
| 373 | //
|
---|
| 374 | const uInt nRows = on->nRow();
|
---|
| 375 | if (off->nRow() != nRows) {
|
---|
| 376 | throw (AipsError("Input Scan Tables must have the same number of rows"));
|
---|
| 377 | }
|
---|
[85] | 378 |
|
---|
[130] | 379 | // Input Tables and columns
|
---|
| 380 |
|
---|
[9] | 381 | Table ton = on->table();
|
---|
| 382 | Table toff = off->table();
|
---|
[85] | 383 | ROArrayColumn<Float> tsys(toff, "TSYS");
|
---|
[9] | 384 | ROScalarColumn<Double> mjd(ton, "TIME");
|
---|
[15] | 385 | ROScalarColumn<Double> integr(ton, "INTERVAL");
|
---|
[9] | 386 | ROScalarColumn<String> srcn(ton, "SRCNAME");
|
---|
[38] | 387 | ROArrayColumn<uInt> freqidc(ton, "FREQID");
|
---|
| 388 |
|
---|
[130] | 389 | // Output Table cloned from input
|
---|
[85] | 390 |
|
---|
[171] | 391 | SDMemTable* pTabOut = new SDMemTable(*on, True);
|
---|
[130] | 392 |
|
---|
| 393 | // Loop over rows
|
---|
| 394 |
|
---|
| 395 | for (uInt i=0; i<nRows; i++) {
|
---|
| 396 | MaskedArray<Float> mon(on->rowAsMaskedArray(i));
|
---|
| 397 | MaskedArray<Float> moff(off->rowAsMaskedArray(i));
|
---|
| 398 | IPosition ipon = mon.shape();
|
---|
| 399 | IPosition ipoff = moff.shape();
|
---|
| 400 | //
|
---|
| 401 | Array<Float> tsarr;
|
---|
| 402 | tsys.get(i, tsarr);
|
---|
| 403 | if (ipon != ipoff && ipon != tsarr.shape()) {
|
---|
| 404 | throw(AipsError("on/off not conformant"));
|
---|
| 405 | }
|
---|
| 406 |
|
---|
| 407 | // Compute quotient
|
---|
| 408 |
|
---|
| 409 | MaskedArray<Float> tmp = (mon-moff);
|
---|
| 410 | Array<Float> out(tmp.getArray());
|
---|
| 411 | out /= moff;
|
---|
| 412 | out *= tsarr;
|
---|
[163] | 413 | Array<Bool> outflagsb = mon.getMask() && moff.getMask();
|
---|
[130] | 414 |
|
---|
| 415 | // Fill container for this row
|
---|
| 416 |
|
---|
[184] | 417 | SDContainer sc = on->getSDContainer(i);
|
---|
[163] | 418 | //
|
---|
[185] | 419 | putDataInSDC(sc, out, outflagsb);
|
---|
[130] | 420 | sc.putTsys(tsarr);
|
---|
[184] | 421 | sc.scanid = i;
|
---|
[130] | 422 |
|
---|
| 423 | // Put new row in output Table
|
---|
| 424 |
|
---|
[171] | 425 | pTabOut->putSDContainer(sc);
|
---|
[130] | 426 | }
|
---|
| 427 | //
|
---|
[171] | 428 | return CountedPtr<SDMemTable>(pTabOut);
|
---|
[9] | 429 | }
|
---|
[48] | 430 |
|
---|
[146] | 431 |
|
---|
| 432 |
|
---|
[185] | 433 | std::vector<float> SDMath::statistic(const CountedPtr<SDMemTable>& in,
|
---|
| 434 | const std::vector<bool>& mask,
|
---|
| 435 | const String& which)
|
---|
[130] | 436 | //
|
---|
| 437 | // Perhaps iteration over pol/beam/if should be in here
|
---|
| 438 | // and inside the nrow iteration ?
|
---|
| 439 | //
|
---|
| 440 | {
|
---|
| 441 | const uInt nRow = in->nRow();
|
---|
| 442 | std::vector<float> result(nRow);
|
---|
| 443 | Vector<Bool> msk(mask);
|
---|
| 444 |
|
---|
| 445 | // Specify cursor location
|
---|
| 446 |
|
---|
[152] | 447 | IPosition start, end;
|
---|
[185] | 448 | getCursorLocation(start, end, *in);
|
---|
[130] | 449 |
|
---|
| 450 | // Loop over rows
|
---|
| 451 |
|
---|
| 452 | const uInt nEl = msk.nelements();
|
---|
| 453 | for (uInt ii=0; ii < in->nRow(); ++ii) {
|
---|
| 454 |
|
---|
| 455 | // Get row and deconstruct
|
---|
| 456 |
|
---|
| 457 | MaskedArray<Float> marr(in->rowAsMaskedArray(ii));
|
---|
| 458 | Array<Float> arr = marr.getArray();
|
---|
| 459 | Array<Bool> barr = marr.getMask();
|
---|
| 460 |
|
---|
| 461 | // Access desired piece of data
|
---|
| 462 |
|
---|
| 463 | Array<Float> v((arr(start,end)).nonDegenerate());
|
---|
| 464 | Array<Bool> m((barr(start,end)).nonDegenerate());
|
---|
| 465 |
|
---|
| 466 | // Apply OTF mask
|
---|
| 467 |
|
---|
| 468 | MaskedArray<Float> tmp;
|
---|
| 469 | if (m.nelements()==nEl) {
|
---|
| 470 | tmp.setData(v,m&&msk);
|
---|
| 471 | } else {
|
---|
| 472 | tmp.setData(v,m);
|
---|
| 473 | }
|
---|
| 474 |
|
---|
| 475 | // Get statistic
|
---|
| 476 |
|
---|
[144] | 477 | result[ii] = mathutil::statistics(which, tmp);
|
---|
[130] | 478 | }
|
---|
| 479 | //
|
---|
| 480 | return result;
|
---|
| 481 | }
|
---|
| 482 |
|
---|
[146] | 483 |
|
---|
[185] | 484 | SDMemTable* SDMath::bin(const SDMemTable& in, Int width)
|
---|
[144] | 485 | {
|
---|
[169] | 486 | SDHeader sh = in.getSDHeader();
|
---|
| 487 | SDMemTable* pTabOut = new SDMemTable(in, True);
|
---|
[163] | 488 |
|
---|
[169] | 489 | // Bin up SpectralCoordinates
|
---|
[163] | 490 |
|
---|
[169] | 491 | IPosition factors(1);
|
---|
| 492 | factors(0) = width;
|
---|
| 493 | for (uInt j=0; j<in.nCoordinates(); ++j) {
|
---|
| 494 | CoordinateSystem cSys;
|
---|
| 495 | cSys.addCoordinate(in.getCoordinate(j));
|
---|
| 496 | CoordinateSystem cSysBin =
|
---|
[185] | 497 | CoordinateUtil::makeBinnedCoordinateSystem(factors, cSys, False);
|
---|
[169] | 498 | //
|
---|
| 499 | SpectralCoordinate sCBin = cSysBin.spectralCoordinate(0);
|
---|
| 500 | pTabOut->setCoordinate(sCBin, j);
|
---|
| 501 | }
|
---|
[163] | 502 |
|
---|
[169] | 503 | // Use RebinLattice to find shape
|
---|
[130] | 504 |
|
---|
[169] | 505 | IPosition shapeIn(1,sh.nchan);
|
---|
[185] | 506 | IPosition shapeOut = RebinLattice<Float>::rebinShape(shapeIn, factors);
|
---|
[169] | 507 | sh.nchan = shapeOut(0);
|
---|
| 508 | pTabOut->putSDHeader(sh);
|
---|
[144] | 509 |
|
---|
| 510 |
|
---|
[169] | 511 | // Loop over rows and bin along channel axis
|
---|
| 512 |
|
---|
| 513 | for (uInt i=0; i < in.nRow(); ++i) {
|
---|
| 514 | SDContainer sc = in.getSDContainer(i);
|
---|
[144] | 515 | //
|
---|
[169] | 516 | Array<Float> tSys(sc.getTsys()); // Get it out before sc changes shape
|
---|
[144] | 517 |
|
---|
[169] | 518 | // Bin up spectrum
|
---|
[144] | 519 |
|
---|
[169] | 520 | MaskedArray<Float> marr(in.rowAsMaskedArray(i));
|
---|
| 521 | MaskedArray<Float> marrout;
|
---|
[221] | 522 | LatticeUtilities::bin(marrout, marr, asap::ChanAxis, width);
|
---|
[144] | 523 |
|
---|
[169] | 524 | // Put back the binned data and flags
|
---|
[144] | 525 |
|
---|
[169] | 526 | IPosition ip2 = marrout.shape();
|
---|
| 527 | sc.resize(ip2);
|
---|
[146] | 528 | //
|
---|
[185] | 529 | putDataInSDC(sc, marrout.getArray(), marrout.getMask());
|
---|
[146] | 530 |
|
---|
[169] | 531 | // Bin up Tsys.
|
---|
[146] | 532 |
|
---|
[169] | 533 | Array<Bool> allGood(tSys.shape(),True);
|
---|
| 534 | MaskedArray<Float> tSysIn(tSys, allGood, True);
|
---|
[146] | 535 | //
|
---|
[169] | 536 | MaskedArray<Float> tSysOut;
|
---|
[221] | 537 | LatticeUtilities::bin(tSysOut, tSysIn, asap::ChanAxis, width);
|
---|
[169] | 538 | sc.putTsys(tSysOut.getArray());
|
---|
[146] | 539 | //
|
---|
[169] | 540 | pTabOut->putSDContainer(sc);
|
---|
| 541 | }
|
---|
| 542 | return pTabOut;
|
---|
[146] | 543 | }
|
---|
| 544 |
|
---|
[185] | 545 | SDMemTable* SDMath::simpleOperate(const SDMemTable& in, Float val, Bool doAll,
|
---|
| 546 | uInt what)
|
---|
[152] | 547 | //
|
---|
| 548 | // what = 0 Multiply
|
---|
| 549 | // 1 Add
|
---|
[146] | 550 | {
|
---|
[152] | 551 | SDMemTable* pOut = new SDMemTable(in,False);
|
---|
| 552 | const Table& tOut = pOut->table();
|
---|
| 553 | ArrayColumn<Float> spec(tOut,"SPECTRA");
|
---|
[146] | 554 | //
|
---|
[152] | 555 | if (doAll) {
|
---|
| 556 | for (uInt i=0; i < tOut.nrow(); i++) {
|
---|
| 557 |
|
---|
| 558 | // Get
|
---|
| 559 |
|
---|
| 560 | MaskedArray<Float> marr(pOut->rowAsMaskedArray(i));
|
---|
| 561 |
|
---|
| 562 | // Operate
|
---|
| 563 |
|
---|
| 564 | if (what==0) {
|
---|
| 565 | marr *= val;
|
---|
| 566 | } else if (what==1) {
|
---|
| 567 | marr += val;
|
---|
| 568 | }
|
---|
| 569 |
|
---|
| 570 | // Put
|
---|
| 571 |
|
---|
| 572 | spec.put(i, marr.getArray());
|
---|
| 573 | }
|
---|
| 574 | } else {
|
---|
| 575 |
|
---|
| 576 | // Get cursor location
|
---|
| 577 |
|
---|
| 578 | IPosition start, end;
|
---|
[185] | 579 | getCursorLocation(start, end, in);
|
---|
[152] | 580 | //
|
---|
| 581 | for (uInt i=0; i < tOut.nrow(); i++) {
|
---|
| 582 |
|
---|
| 583 | // Get
|
---|
| 584 |
|
---|
| 585 | MaskedArray<Float> dataIn(pOut->rowAsMaskedArray(i));
|
---|
| 586 |
|
---|
| 587 | // Modify. More work than we would like to deal with the mask
|
---|
| 588 |
|
---|
| 589 | Array<Float>& values = dataIn.getRWArray();
|
---|
| 590 | Array<Bool> mask(dataIn.getMask());
|
---|
| 591 | //
|
---|
| 592 | Array<Float> values2 = values(start,end);
|
---|
| 593 | Array<Bool> mask2 = mask(start,end);
|
---|
| 594 | MaskedArray<Float> t(values2,mask2);
|
---|
| 595 | if (what==0) {
|
---|
| 596 | t *= val;
|
---|
| 597 | } else if (what==1) {
|
---|
| 598 | t += val;
|
---|
| 599 | }
|
---|
| 600 | values(start, end) = t.getArray(); // Write back into 'dataIn'
|
---|
| 601 |
|
---|
| 602 | // Put
|
---|
| 603 | spec.put(i, dataIn.getArray());
|
---|
| 604 | }
|
---|
| 605 | }
|
---|
| 606 | //
|
---|
[146] | 607 | return pOut;
|
---|
| 608 | }
|
---|
| 609 |
|
---|
| 610 |
|
---|
[152] | 611 |
|
---|
[185] | 612 | SDMemTable* SDMath::averagePol(const SDMemTable& in, const Vector<Bool>& mask)
|
---|
[152] | 613 | //
|
---|
[165] | 614 | // Average all polarizations together, weighted by variance
|
---|
| 615 | //
|
---|
| 616 | {
|
---|
| 617 | // WeightType wtType = NONE;
|
---|
[185] | 618 | // convertWeightString(wtType, weight);
|
---|
[165] | 619 |
|
---|
| 620 | const uInt nRows = in.nRow();
|
---|
[209] | 621 | const uInt polAxis = asap::PolAxis; // Polarization axis
|
---|
| 622 | const uInt chanAxis = asap::ChanAxis; // Spectrum axis
|
---|
[165] | 623 |
|
---|
| 624 | // Create output Table and reshape number of polarizations
|
---|
| 625 |
|
---|
| 626 | Bool clear=True;
|
---|
| 627 | SDMemTable* pTabOut = new SDMemTable(in, clear);
|
---|
| 628 | SDHeader header = pTabOut->getSDHeader();
|
---|
| 629 | header.npol = 1;
|
---|
| 630 | pTabOut->putSDHeader(header);
|
---|
| 631 |
|
---|
| 632 | // Shape of input and output data
|
---|
| 633 |
|
---|
| 634 | const IPosition& shapeIn = in.rowAsMaskedArray(0u, False).shape();
|
---|
| 635 | IPosition shapeOut(shapeIn);
|
---|
| 636 | shapeOut(polAxis) = 1; // Average all polarizations
|
---|
| 637 | //
|
---|
| 638 | const uInt nChan = shapeIn(chanAxis);
|
---|
| 639 | const IPosition vecShapeOut(4,1,1,1,nChan); // A multi-dim form of a Vector shape
|
---|
| 640 | IPosition start(4), end(4);
|
---|
| 641 |
|
---|
| 642 | // Output arrays
|
---|
| 643 |
|
---|
| 644 | Array<Float> outData(shapeOut, 0.0);
|
---|
| 645 | Array<Bool> outMask(shapeOut, True);
|
---|
| 646 | const IPosition axes(2, 2, 3); // pol-channel plane
|
---|
| 647 | //
|
---|
| 648 | const Bool useMask = (mask.nelements() == shapeIn(chanAxis));
|
---|
| 649 |
|
---|
| 650 | // Loop over rows
|
---|
| 651 |
|
---|
| 652 | for (uInt iRow=0; iRow<nRows; iRow++) {
|
---|
| 653 |
|
---|
| 654 | // Get data for this row
|
---|
| 655 |
|
---|
| 656 | MaskedArray<Float> marr(in.rowAsMaskedArray(iRow));
|
---|
| 657 | Array<Float>& arr = marr.getRWArray();
|
---|
| 658 | const Array<Bool>& barr = marr.getMask();
|
---|
| 659 |
|
---|
| 660 | // Make iterators to iterate by pol-channel planes
|
---|
| 661 |
|
---|
| 662 | ReadOnlyArrayIterator<Float> itDataPlane(arr, axes);
|
---|
| 663 | ReadOnlyArrayIterator<Bool> itMaskPlane(barr, axes);
|
---|
| 664 |
|
---|
| 665 | // Accumulations
|
---|
| 666 |
|
---|
| 667 | Float fac = 1.0;
|
---|
| 668 | Vector<Float> vecSum(nChan,0.0);
|
---|
| 669 |
|
---|
| 670 | // Iterate through data by pol-channel planes
|
---|
| 671 |
|
---|
| 672 | while (!itDataPlane.pastEnd()) {
|
---|
| 673 |
|
---|
| 674 | // Iterate through plane by polarization and accumulate Vectors
|
---|
| 675 |
|
---|
| 676 | Vector<Float> t1(nChan); t1 = 0.0;
|
---|
| 677 | Vector<Bool> t2(nChan); t2 = True;
|
---|
| 678 | MaskedArray<Float> vecSum(t1,t2);
|
---|
| 679 | Float varSum = 0.0;
|
---|
| 680 | {
|
---|
| 681 | ReadOnlyVectorIterator<Float> itDataVec(itDataPlane.array(), 1);
|
---|
| 682 | ReadOnlyVectorIterator<Bool> itMaskVec(itMaskPlane.array(), 1);
|
---|
| 683 | while (!itDataVec.pastEnd()) {
|
---|
| 684 |
|
---|
| 685 | // Create MA of data & mask (optionally including OTF mask) and get variance
|
---|
| 686 |
|
---|
| 687 | if (useMask) {
|
---|
| 688 | const MaskedArray<Float> spec(itDataVec.vector(),mask&&itMaskVec.vector());
|
---|
| 689 | fac = 1.0 / variance(spec);
|
---|
| 690 | } else {
|
---|
| 691 | const MaskedArray<Float> spec(itDataVec.vector(),itMaskVec.vector());
|
---|
| 692 | fac = 1.0 / variance(spec);
|
---|
| 693 | }
|
---|
| 694 |
|
---|
| 695 | // Normalize spectrum (without OTF mask) and accumulate
|
---|
| 696 |
|
---|
| 697 | const MaskedArray<Float> spec(fac*itDataVec.vector(), itMaskVec.vector());
|
---|
| 698 | vecSum += spec;
|
---|
| 699 | varSum += fac;
|
---|
| 700 |
|
---|
| 701 | // Next
|
---|
| 702 |
|
---|
| 703 | itDataVec.next();
|
---|
| 704 | itMaskVec.next();
|
---|
| 705 | }
|
---|
| 706 | }
|
---|
| 707 |
|
---|
| 708 | // Normalize summed spectrum
|
---|
| 709 |
|
---|
| 710 | vecSum /= varSum;
|
---|
| 711 |
|
---|
| 712 | // FInd position in input data array. We are iterating by pol-channel
|
---|
| 713 | // plane so all that will change is beam and IF and that's what we want.
|
---|
| 714 |
|
---|
| 715 | IPosition pos = itDataPlane.pos();
|
---|
| 716 |
|
---|
| 717 | // Write out data. This is a bit messy. We have to reform the Vector
|
---|
| 718 | // accumulator into an Array of shape (1,1,1,nChan)
|
---|
| 719 |
|
---|
| 720 | start = pos;
|
---|
| 721 | end = pos;
|
---|
| 722 | end(chanAxis) = nChan-1;
|
---|
| 723 | outData(start,end) = vecSum.getArray().reform(vecShapeOut);
|
---|
| 724 | outMask(start,end) = vecSum.getMask().reform(vecShapeOut);
|
---|
| 725 |
|
---|
| 726 | // Step to next beam/IF combination
|
---|
| 727 |
|
---|
| 728 | itDataPlane.next();
|
---|
| 729 | itMaskPlane.next();
|
---|
| 730 | }
|
---|
| 731 |
|
---|
| 732 | // Generate output container and write it to output table
|
---|
| 733 |
|
---|
| 734 | SDContainer sc = in.getSDContainer();
|
---|
| 735 | sc.resize(shapeOut);
|
---|
| 736 | //
|
---|
[185] | 737 | putDataInSDC(sc, outData, outMask);
|
---|
[165] | 738 | pTabOut->putSDContainer(sc);
|
---|
| 739 | }
|
---|
| 740 | //
|
---|
| 741 | return pTabOut;
|
---|
| 742 | }
|
---|
[167] | 743 |
|
---|
[169] | 744 |
|
---|
[185] | 745 | SDMemTable* SDMath::smooth(const SDMemTable& in,
|
---|
| 746 | const casa::String& kernelType,
|
---|
| 747 | casa::Float width, Bool doAll)
|
---|
[177] | 748 | {
|
---|
[169] | 749 |
|
---|
[177] | 750 | // Number of channels
|
---|
[169] | 751 |
|
---|
[209] | 752 | const uInt chanAxis = asap::ChanAxis; // Spectral axis
|
---|
[177] | 753 | SDHeader sh = in.getSDHeader();
|
---|
| 754 | const uInt nChan = sh.nchan;
|
---|
| 755 |
|
---|
| 756 | // Generate Kernel
|
---|
| 757 |
|
---|
[185] | 758 | VectorKernel::KernelTypes type = VectorKernel::toKernelType(kernelType);
|
---|
[177] | 759 | Vector<Float> kernel = VectorKernel::make(type, width, nChan, True, False);
|
---|
| 760 |
|
---|
| 761 | // Generate Convolver
|
---|
| 762 |
|
---|
| 763 | IPosition shape(1,nChan);
|
---|
| 764 | Convolver<Float> conv(kernel, shape);
|
---|
| 765 |
|
---|
| 766 | // New Table
|
---|
| 767 |
|
---|
| 768 | SDMemTable* pTabOut = new SDMemTable(in,True);
|
---|
| 769 |
|
---|
| 770 | // Get cursor location
|
---|
| 771 |
|
---|
| 772 | IPosition start, end;
|
---|
[185] | 773 | getCursorLocation(start, end, in);
|
---|
[177] | 774 | //
|
---|
| 775 | IPosition shapeOut(4,1);
|
---|
| 776 |
|
---|
| 777 | // Output Vectors
|
---|
| 778 |
|
---|
| 779 | Vector<Float> valuesOut(nChan);
|
---|
| 780 | Vector<Bool> maskOut(nChan);
|
---|
| 781 |
|
---|
| 782 | // Loop over rows in Table
|
---|
| 783 |
|
---|
| 784 | for (uInt ri=0; ri < in.nRow(); ++ri) {
|
---|
| 785 |
|
---|
| 786 | // Get copy of data
|
---|
| 787 |
|
---|
| 788 | const MaskedArray<Float>& dataIn(in.rowAsMaskedArray(ri));
|
---|
| 789 | AlwaysAssert(dataIn.shape()(chanAxis)==nChan, AipsError);
|
---|
| 790 | //
|
---|
| 791 | Array<Float> valuesIn = dataIn.getArray();
|
---|
| 792 | Array<Bool> maskIn = dataIn.getMask();
|
---|
| 793 |
|
---|
| 794 | // Branch depending on whether we smooth all locations or just
|
---|
| 795 | // those pointed at by the current selection cursor
|
---|
| 796 |
|
---|
| 797 | if (doAll) {
|
---|
[221] | 798 | uInt axis = asap::ChanAxis;
|
---|
[177] | 799 | VectorIterator<Float> itValues(valuesIn, axis);
|
---|
| 800 | VectorIterator<Bool> itMask(maskIn, axis);
|
---|
| 801 | while (!itValues.pastEnd()) {
|
---|
| 802 |
|
---|
| 803 | // Smooth
|
---|
| 804 | if (kernelType==VectorKernel::HANNING) {
|
---|
| 805 | mathutil::hanning(valuesOut, maskOut, itValues.vector(), itMask.vector());
|
---|
| 806 | itMask.vector() = maskOut;
|
---|
| 807 | } else {
|
---|
| 808 | mathutil::replaceMaskByZero(itValues.vector(), itMask.vector());
|
---|
| 809 | conv.linearConv(valuesOut, itValues.vector());
|
---|
| 810 | }
|
---|
| 811 | //
|
---|
| 812 | itValues.vector() = valuesOut;
|
---|
| 813 | //
|
---|
| 814 | itValues.next();
|
---|
| 815 | itMask.next();
|
---|
| 816 | }
|
---|
| 817 | } else {
|
---|
| 818 |
|
---|
| 819 | // Set multi-dim Vector shape
|
---|
| 820 |
|
---|
| 821 | shapeOut(chanAxis) = valuesIn.shape()(chanAxis);
|
---|
| 822 |
|
---|
| 823 | // Stuff about with shapes so that we don't have conformance run-time errors
|
---|
| 824 |
|
---|
| 825 | Vector<Float> valuesIn2 = valuesIn(start,end).nonDegenerate();
|
---|
| 826 | Vector<Bool> maskIn2 = maskIn(start,end).nonDegenerate();
|
---|
| 827 |
|
---|
| 828 | // Smooth
|
---|
| 829 |
|
---|
| 830 | if (kernelType==VectorKernel::HANNING) {
|
---|
| 831 | mathutil::hanning(valuesOut, maskOut, valuesIn2, maskIn2);
|
---|
| 832 | maskIn(start,end) = maskOut.reform(shapeOut);
|
---|
| 833 | } else {
|
---|
| 834 | mathutil::replaceMaskByZero(valuesIn2, maskIn2);
|
---|
| 835 | conv.linearConv(valuesOut, valuesIn2);
|
---|
| 836 | }
|
---|
| 837 | //
|
---|
| 838 | valuesIn(start,end) = valuesOut.reform(shapeOut);
|
---|
| 839 | }
|
---|
| 840 |
|
---|
| 841 | // Create and put back
|
---|
| 842 |
|
---|
| 843 | SDContainer sc = in.getSDContainer(ri);
|
---|
[185] | 844 | putDataInSDC(sc, valuesIn, maskIn);
|
---|
[177] | 845 | //
|
---|
| 846 | pTabOut->putSDContainer(sc);
|
---|
| 847 | }
|
---|
| 848 | //
|
---|
| 849 | return pTabOut;
|
---|
| 850 | }
|
---|
| 851 |
|
---|
| 852 |
|
---|
[221] | 853 | SDMemTable* SDMath::convertFlux (const SDMemTable& in, Float a, Float eta, Bool doAll)
|
---|
| 854 | //
|
---|
| 855 | // As it is, this function could be implemented with 'simpleOperate'
|
---|
| 856 | // However, I anticipate that eventually we will look the conversion
|
---|
| 857 | // values up in a Table and apply them in a frequency dependent way,
|
---|
| 858 | // so I have implemented it fully here
|
---|
| 859 | //
|
---|
| 860 | {
|
---|
| 861 | SDHeader sh = in.getSDHeader();
|
---|
| 862 | SDMemTable* pTabOut = new SDMemTable(in, True);
|
---|
[177] | 863 |
|
---|
[221] | 864 | // FInd out how to convert values into Jy and K (e.g. units might be mJy or mK)
|
---|
| 865 | // Also automatically find out what we are converting to according to the
|
---|
| 866 | // flux unit
|
---|
[177] | 867 |
|
---|
[221] | 868 | Unit fluxUnit(sh.fluxunit);
|
---|
| 869 | Unit K(String("K"));
|
---|
| 870 | Unit JY(String("Jy"));
|
---|
| 871 | //
|
---|
| 872 | Bool toKelvin = True;
|
---|
| 873 | Double inFac = 1.0;
|
---|
| 874 | if (fluxUnit==JY) {
|
---|
| 875 | cerr << "Converting to K" << endl;
|
---|
| 876 | //
|
---|
| 877 | Quantum<Double> t(1.0,fluxUnit);
|
---|
| 878 | Quantum<Double> t2 = t.get(JY);
|
---|
| 879 | inFac = (t2 / t).getValue();
|
---|
| 880 | //
|
---|
| 881 | toKelvin = True;
|
---|
| 882 | sh.fluxunit = "K";
|
---|
| 883 | } else if (fluxUnit==K) {
|
---|
| 884 | cerr << "Converting to Jy" << endl;
|
---|
| 885 | //
|
---|
| 886 | Quantum<Double> t(1.0,fluxUnit);
|
---|
| 887 | Quantum<Double> t2 = t.get(K);
|
---|
| 888 | inFac = (t2 / t).getValue();
|
---|
| 889 | //
|
---|
| 890 | toKelvin = False;
|
---|
| 891 | sh.fluxunit = "Jy";
|
---|
| 892 | } else {
|
---|
| 893 | throw AipsError("Unrecognized brightness units in Table - must be consistent with Jy or K");
|
---|
| 894 | }
|
---|
| 895 | pTabOut->putSDHeader(sh);
|
---|
[177] | 896 |
|
---|
[221] | 897 | // Compute conversion factor. 'a' and 'eta' are really frequency, time and
|
---|
| 898 | // telescope dependent and should be looked// up in a table
|
---|
| 899 |
|
---|
| 900 | Float factor = 2.0 * inFac * 1.0e-7 * 1.0e26 * QC::k.getValue(Unit(String("erg/K"))) / a / eta;
|
---|
| 901 | if (toKelvin) {
|
---|
| 902 | factor = 1.0 / factor;
|
---|
| 903 | }
|
---|
| 904 | cerr << "Applying conversion factor = " << factor << endl;
|
---|
| 905 |
|
---|
| 906 | // For operations only on specified cursor location
|
---|
| 907 |
|
---|
| 908 | IPosition start, end;
|
---|
| 909 | getCursorLocation(start, end, in);
|
---|
| 910 |
|
---|
| 911 | // Loop over rows and apply factor to spectra
|
---|
| 912 |
|
---|
| 913 | const uInt axis = asap::ChanAxis;
|
---|
| 914 | for (uInt i=0; i < in.nRow(); ++i) {
|
---|
| 915 |
|
---|
| 916 | // Get data
|
---|
| 917 |
|
---|
| 918 | MaskedArray<Float> dataIn(in.rowAsMaskedArray(i));
|
---|
| 919 | Array<Float>& valuesIn = dataIn.getRWArray(); // writable reference
|
---|
| 920 | const Array<Bool>& maskIn = dataIn.getMask();
|
---|
| 921 |
|
---|
| 922 | // Need to apply correct conversion factor (frequency and time dependent)
|
---|
| 923 | // which should be sourced from a Table. For now we just apply the given
|
---|
| 924 | // factor to everything
|
---|
| 925 |
|
---|
| 926 | if (doAll) {
|
---|
| 927 | VectorIterator<Float> itValues(valuesIn, asap::ChanAxis);
|
---|
| 928 | while (!itValues.pastEnd()) {
|
---|
| 929 | itValues.vector() *= factor; // Writes back into dataIn
|
---|
| 930 | //
|
---|
| 931 | itValues.next();
|
---|
| 932 | }
|
---|
| 933 | } else {
|
---|
| 934 | Array<Float> valuesIn2 = valuesIn(start,end);
|
---|
| 935 | valuesIn2 *= factor;
|
---|
| 936 | valuesIn(start,end) = valuesIn2;
|
---|
| 937 | }
|
---|
| 938 |
|
---|
| 939 | // Write out
|
---|
| 940 |
|
---|
| 941 | SDContainer sc = in.getSDContainer(i);
|
---|
| 942 | putDataInSDC(sc, valuesIn, maskIn);
|
---|
| 943 | //
|
---|
| 944 | pTabOut->putSDContainer(sc);
|
---|
| 945 | }
|
---|
| 946 | return pTabOut;
|
---|
| 947 | }
|
---|
| 948 |
|
---|
| 949 |
|
---|
| 950 |
|
---|
[227] | 951 | SDMemTable* SDMath::gainElevation (const SDMemTable& in, const String& fileName,
|
---|
| 952 | const String& methodStr, Bool doAll)
|
---|
| 953 | {
|
---|
| 954 | SDHeader sh = in.getSDHeader();
|
---|
| 955 | SDMemTable* pTabOut = new SDMemTable(in, True);
|
---|
| 956 | const uInt nRow = in.nRow();
|
---|
| 957 |
|
---|
[230] | 958 | // Get elevation from SDMemTable data
|
---|
[227] | 959 |
|
---|
| 960 | const Table& tab = in.table();
|
---|
| 961 | ROScalarColumn<Float> elev(tab, "ELEVATION");
|
---|
| 962 | Vector<Float> xOut = elev.getColumn();
|
---|
| 963 | xOut *= Float(180 / C::pi);
|
---|
| 964 | //
|
---|
[230] | 965 | String col0("ELEVATION");
|
---|
| 966 | String col1("FACTOR");
|
---|
[227] | 967 | //
|
---|
[230] | 968 | return correctFromAsciiTable (pTabOut, in, fileName, col0, col1, methodStr, doAll, xOut);
|
---|
| 969 | }
|
---|
[227] | 970 |
|
---|
[230] | 971 |
|
---|
[227] | 972 |
|
---|
| 973 |
|
---|
| 974 |
|
---|
[169] | 975 | // 'private' functions
|
---|
| 976 |
|
---|
[185] | 977 | void SDMath::fillSDC(SDContainer& sc,
|
---|
| 978 | const Array<Bool>& mask,
|
---|
| 979 | const Array<Float>& data,
|
---|
| 980 | const Array<Float>& tSys,
|
---|
| 981 | Int scanID, Double timeStamp,
|
---|
| 982 | Double interval, const String& sourceName,
|
---|
[227] | 983 | const Vector<uInt>& freqID) const
|
---|
[167] | 984 | {
|
---|
[169] | 985 | // Data and mask
|
---|
[167] | 986 |
|
---|
[185] | 987 | putDataInSDC(sc, data, mask);
|
---|
[167] | 988 |
|
---|
[169] | 989 | // TSys
|
---|
| 990 |
|
---|
| 991 | sc.putTsys(tSys);
|
---|
| 992 |
|
---|
| 993 | // Time things
|
---|
| 994 |
|
---|
| 995 | sc.timestamp = timeStamp;
|
---|
| 996 | sc.interval = interval;
|
---|
| 997 | sc.scanid = scanID;
|
---|
[167] | 998 | //
|
---|
[169] | 999 | sc.sourcename = sourceName;
|
---|
| 1000 | sc.putFreqMap(freqID);
|
---|
| 1001 | }
|
---|
[167] | 1002 |
|
---|
[185] | 1003 | void SDMath::normalize(MaskedArray<Float>& sum,
|
---|
[169] | 1004 | const Array<Float>& sumSq,
|
---|
| 1005 | const Array<Float>& nPts,
|
---|
| 1006 | WeightType wtType, Int axis,
|
---|
[227] | 1007 | Int nAxesSub) const
|
---|
[169] | 1008 | {
|
---|
| 1009 | IPosition pos2(nAxesSub,0);
|
---|
| 1010 | //
|
---|
| 1011 | if (wtType==NONE) {
|
---|
[167] | 1012 |
|
---|
[169] | 1013 | // We just average by the number of points accumulated.
|
---|
| 1014 | // We need to make a MA out of nPts so that no divide by
|
---|
| 1015 | // zeros occur
|
---|
[167] | 1016 |
|
---|
[169] | 1017 | MaskedArray<Float> t(nPts, (nPts>Float(0.0)));
|
---|
| 1018 | sum /= t;
|
---|
| 1019 | } else if (wtType==VAR) {
|
---|
[167] | 1020 |
|
---|
[169] | 1021 | // Normalize each spectrum by sum(1/var) where the variance
|
---|
| 1022 | // is worked out for each spectrum
|
---|
| 1023 |
|
---|
| 1024 | Array<Float>& data = sum.getRWArray();
|
---|
| 1025 | VectorIterator<Float> itData(data, axis);
|
---|
| 1026 | while (!itData.pastEnd()) {
|
---|
| 1027 | pos2 = itData.pos().getFirst(nAxesSub);
|
---|
| 1028 | itData.vector() /= sumSq(pos2);
|
---|
| 1029 | itData.next();
|
---|
| 1030 | }
|
---|
| 1031 | } else if (wtType==TSYS) {
|
---|
| 1032 | }
|
---|
| 1033 | }
|
---|
| 1034 |
|
---|
| 1035 |
|
---|
[185] | 1036 | void SDMath::accumulate(Double& timeSum, Double& intSum, Int& nAccum,
|
---|
| 1037 | MaskedArray<Float>& sum, Array<Float>& sumSq,
|
---|
| 1038 | Array<Float>& nPts, Array<Float>& tSysSum,
|
---|
| 1039 | const Array<Float>& tSys, const Array<Float>& nInc,
|
---|
| 1040 | const Vector<Bool>& mask, Double time, Double interval,
|
---|
| 1041 | const Block<CountedPtr<SDMemTable> >& in,
|
---|
| 1042 | uInt iTab, uInt iRow, uInt axis,
|
---|
| 1043 | uInt nAxesSub, Bool useMask,
|
---|
[227] | 1044 | WeightType wtType) const
|
---|
[169] | 1045 | {
|
---|
| 1046 |
|
---|
| 1047 | // Get data
|
---|
| 1048 |
|
---|
| 1049 | MaskedArray<Float> dataIn(in[iTab]->rowAsMaskedArray(iRow));
|
---|
| 1050 | Array<Float>& valuesIn = dataIn.getRWArray(); // writable reference
|
---|
| 1051 | const Array<Bool>& maskIn = dataIn.getMask(); // RO reference
|
---|
[167] | 1052 | //
|
---|
[169] | 1053 | if (wtType==NONE) {
|
---|
| 1054 | const MaskedArray<Float> n(nInc,dataIn.getMask());
|
---|
| 1055 | nPts += n; // Only accumulates where mask==T
|
---|
| 1056 | } else if (wtType==VAR) {
|
---|
[167] | 1057 |
|
---|
[169] | 1058 | // We are going to average the data, weighted by the noise for each pol, beam and IF.
|
---|
| 1059 | // So therefore we need to iterate through by spectrum (axis 3)
|
---|
[167] | 1060 |
|
---|
[169] | 1061 | VectorIterator<Float> itData(valuesIn, axis);
|
---|
| 1062 | ReadOnlyVectorIterator<Bool> itMask(maskIn, axis);
|
---|
| 1063 | Float fac = 1.0;
|
---|
| 1064 | IPosition pos(nAxesSub,0);
|
---|
| 1065 | //
|
---|
| 1066 | while (!itData.pastEnd()) {
|
---|
[167] | 1067 |
|
---|
[169] | 1068 | // Make MaskedArray of Vector, optionally apply OTF mask, and find scaling factor
|
---|
[167] | 1069 |
|
---|
[169] | 1070 | if (useMask) {
|
---|
| 1071 | MaskedArray<Float> tmp(itData.vector(),mask&&itMask.vector());
|
---|
| 1072 | fac = 1.0/variance(tmp);
|
---|
| 1073 | } else {
|
---|
| 1074 | MaskedArray<Float> tmp(itData.vector(),itMask.vector());
|
---|
| 1075 | fac = 1.0/variance(tmp);
|
---|
| 1076 | }
|
---|
| 1077 |
|
---|
| 1078 | // Scale data
|
---|
| 1079 |
|
---|
| 1080 | itData.vector() *= fac; // Writes back into 'dataIn'
|
---|
[167] | 1081 | //
|
---|
[169] | 1082 | // Accumulate variance per if/pol/beam averaged over spectrum
|
---|
| 1083 | // This method to get pos2 from itData.pos() is only valid
|
---|
| 1084 | // because the spectral axis is the last one (so we can just
|
---|
| 1085 | // copy the first nAXesSub positions out)
|
---|
[167] | 1086 |
|
---|
[169] | 1087 | pos = itData.pos().getFirst(nAxesSub);
|
---|
| 1088 | sumSq(pos) += fac;
|
---|
| 1089 | //
|
---|
| 1090 | itData.next();
|
---|
| 1091 | itMask.next();
|
---|
| 1092 | }
|
---|
| 1093 | } else if (wtType==TSYS) {
|
---|
| 1094 | }
|
---|
[167] | 1095 |
|
---|
[169] | 1096 | // Accumulate sum of (possibly scaled) data
|
---|
| 1097 |
|
---|
| 1098 | sum += dataIn;
|
---|
| 1099 |
|
---|
| 1100 | // Accumulate Tsys, time, and interval
|
---|
| 1101 |
|
---|
| 1102 | tSysSum += tSys;
|
---|
| 1103 | timeSum += time;
|
---|
| 1104 | intSum += interval;
|
---|
| 1105 | nAccum += 1;
|
---|
| 1106 | }
|
---|
| 1107 |
|
---|
| 1108 |
|
---|
| 1109 |
|
---|
| 1110 |
|
---|
[185] | 1111 | void SDMath::getCursorLocation(IPosition& start, IPosition& end,
|
---|
[227] | 1112 | const SDMemTable& in) const
|
---|
[169] | 1113 | {
|
---|
| 1114 | const uInt nDim = 4;
|
---|
| 1115 | const uInt i = in.getBeam();
|
---|
| 1116 | const uInt j = in.getIF();
|
---|
| 1117 | const uInt k = in.getPol();
|
---|
| 1118 | const uInt n = in.nChan();
|
---|
[167] | 1119 | //
|
---|
[169] | 1120 | start.resize(nDim);
|
---|
| 1121 | start(0) = i;
|
---|
| 1122 | start(1) = j;
|
---|
| 1123 | start(2) = k;
|
---|
| 1124 | start(3) = 0;
|
---|
[167] | 1125 | //
|
---|
[169] | 1126 | end.resize(nDim);
|
---|
| 1127 | end(0) = i;
|
---|
| 1128 | end(1) = j;
|
---|
| 1129 | end(2) = k;
|
---|
| 1130 | end(3) = n-1;
|
---|
| 1131 | }
|
---|
| 1132 |
|
---|
| 1133 |
|
---|
[227] | 1134 | void SDMath::convertWeightString(WeightType& wtType, const String& weightStr) const
|
---|
[169] | 1135 | {
|
---|
| 1136 | String tStr(weightStr);
|
---|
| 1137 | tStr.upcase();
|
---|
| 1138 | if (tStr.contains(String("NONE"))) {
|
---|
| 1139 | wtType = NONE;
|
---|
| 1140 | } else if (tStr.contains(String("VAR"))) {
|
---|
| 1141 | wtType = VAR;
|
---|
| 1142 | } else if (tStr.contains(String("TSYS"))) {
|
---|
| 1143 | wtType = TSYS;
|
---|
[185] | 1144 | throw(AipsError("T_sys weighting not yet implemented"));
|
---|
[169] | 1145 | } else {
|
---|
[185] | 1146 | throw(AipsError("Unrecognized weighting type"));
|
---|
[167] | 1147 | }
|
---|
| 1148 | }
|
---|
| 1149 |
|
---|
[227] | 1150 | void SDMath::convertInterpString(Int& type, const String& interp) const
|
---|
| 1151 | {
|
---|
| 1152 | String tStr(interp);
|
---|
| 1153 | tStr.upcase();
|
---|
| 1154 | if (tStr.contains(String("NEAR"))) {
|
---|
| 1155 | type = InterpolateArray1D<Float,Float>::nearestNeighbour;
|
---|
| 1156 | } else if (tStr.contains(String("LIN"))) {
|
---|
| 1157 | type = InterpolateArray1D<Float,Float>::linear;
|
---|
| 1158 | } else if (tStr.contains(String("CUB"))) {
|
---|
| 1159 | type = InterpolateArray1D<Float,Float>::cubic;
|
---|
| 1160 | } else if (tStr.contains(String("SPL"))) {
|
---|
| 1161 | type = InterpolateArray1D<Float,Float>::spline;
|
---|
| 1162 | } else {
|
---|
| 1163 | throw(AipsError("Unrecognized interpolation type"));
|
---|
| 1164 | }
|
---|
| 1165 | }
|
---|
| 1166 |
|
---|
[185] | 1167 | void SDMath::putDataInSDC(SDContainer& sc, const Array<Float>& data,
|
---|
[227] | 1168 | const Array<Bool>& mask) const
|
---|
[169] | 1169 | {
|
---|
| 1170 | sc.putSpectrum(data);
|
---|
| 1171 | //
|
---|
| 1172 | Array<uChar> outflags(data.shape());
|
---|
| 1173 | convertArray(outflags,!mask);
|
---|
| 1174 | sc.putFlags(outflags);
|
---|
| 1175 | }
|
---|
[227] | 1176 |
|
---|
| 1177 | Table SDMath::readAsciiFile (const String& fileName) const
|
---|
| 1178 | {
|
---|
[230] | 1179 | String formatString;
|
---|
| 1180 | Table tbl = readAsciiTable (formatString, Table::Memory, fileName, "", "", False);
|
---|
[227] | 1181 | return tbl;
|
---|
| 1182 | }
|
---|
[230] | 1183 |
|
---|
| 1184 |
|
---|
| 1185 | SDMemTable* SDMath::correctFromAsciiTable(SDMemTable* pTabOut,
|
---|
| 1186 | const SDMemTable& in, const String& fileName,
|
---|
| 1187 | const String& col0, const String& col1,
|
---|
| 1188 | const String& methodStr, Bool doAll,
|
---|
| 1189 | const Vector<Float>& xOut)
|
---|
| 1190 | {
|
---|
| 1191 |
|
---|
| 1192 | // Read gain-elevation ascii file data into a Table.
|
---|
| 1193 |
|
---|
| 1194 | Table tTable = readAsciiFile (fileName);
|
---|
| 1195 | // tTable.markForDelete();
|
---|
| 1196 | //
|
---|
| 1197 | return correctFromTable (pTabOut, in, tTable, col0, col1, methodStr, doAll, xOut);
|
---|
| 1198 | }
|
---|
| 1199 |
|
---|
| 1200 | SDMemTable* SDMath::correctFromTable(SDMemTable* pTabOut, const SDMemTable& in, const Table& tTable,
|
---|
| 1201 | const String& col0, const String& col1,
|
---|
| 1202 | const String& methodStr, Bool doAll,
|
---|
| 1203 | const Vector<Float>& xOut)
|
---|
| 1204 | {
|
---|
| 1205 |
|
---|
| 1206 | // Get data from Table
|
---|
| 1207 |
|
---|
| 1208 | ROScalarColumn<Float> geElCol(tTable, col0);
|
---|
| 1209 | ROScalarColumn<Float> geFacCol(tTable, col1);
|
---|
| 1210 | Vector<Float> xIn = geElCol.getColumn();
|
---|
| 1211 | Vector<Float> yIn = geFacCol.getColumn();
|
---|
| 1212 | Vector<Bool> maskIn(xIn.nelements(),True);
|
---|
| 1213 |
|
---|
| 1214 | // Interpolate (and extrapolate) with desired method
|
---|
| 1215 |
|
---|
| 1216 | Int method = 0;
|
---|
| 1217 | convertInterpString(method, methodStr);
|
---|
| 1218 | //
|
---|
| 1219 | Vector<Float> yOut;
|
---|
| 1220 | Vector<Bool> maskOut;
|
---|
| 1221 | InterpolateArray1D<Float,Float>::interpolate(yOut, maskOut, xOut,
|
---|
| 1222 | xIn, yIn, maskIn, method,
|
---|
| 1223 | True, True);
|
---|
| 1224 |
|
---|
| 1225 | // For operations only on specified cursor location
|
---|
| 1226 |
|
---|
| 1227 | IPosition start, end;
|
---|
| 1228 | getCursorLocation(start, end, in);
|
---|
| 1229 |
|
---|
| 1230 | // Loop over rows and interpolate correction factor
|
---|
| 1231 |
|
---|
| 1232 | const uInt axis = asap::ChanAxis;
|
---|
| 1233 | for (uInt i=0; i < in.nRow(); ++i) {
|
---|
| 1234 |
|
---|
| 1235 | // Get data
|
---|
| 1236 |
|
---|
| 1237 | MaskedArray<Float> dataIn(in.rowAsMaskedArray(i));
|
---|
| 1238 | Array<Float>& valuesIn = dataIn.getRWArray(); // writable reference
|
---|
| 1239 | const Array<Bool>& maskIn = dataIn.getMask();
|
---|
| 1240 |
|
---|
| 1241 | // Apply factor
|
---|
| 1242 |
|
---|
| 1243 | if (doAll) {
|
---|
| 1244 | VectorIterator<Float> itValues(valuesIn, asap::ChanAxis);
|
---|
| 1245 | while (!itValues.pastEnd()) {
|
---|
| 1246 | itValues.vector() *= yOut(i); // Writes back into dataIn
|
---|
| 1247 | //
|
---|
| 1248 | itValues.next();
|
---|
| 1249 | }
|
---|
| 1250 | } else {
|
---|
| 1251 | Array<Float> valuesIn2 = valuesIn(start,end);
|
---|
| 1252 | valuesIn2 *= yOut(i);
|
---|
| 1253 | valuesIn(start,end) = valuesIn2;
|
---|
| 1254 | }
|
---|
| 1255 |
|
---|
| 1256 | // Write out
|
---|
| 1257 |
|
---|
| 1258 | SDContainer sc = in.getSDContainer(i);
|
---|
| 1259 | putDataInSDC(sc, valuesIn, maskIn);
|
---|
| 1260 | //
|
---|
| 1261 | pTabOut->putSDContainer(sc);
|
---|
| 1262 | }
|
---|
| 1263 | //
|
---|
| 1264 | return pTabOut;
|
---|
| 1265 | }
|
---|
| 1266 |
|
---|