def imposeRF(self,refFrame,year,doy,fid=None):
# convert year doy to fractional year
fyear = pyDate.yeardoy2fyear(year, doy);
# need to translate the 4 char names of
#current npvs to real stnIDs 8 char
stnIdList = list();
translate = self.dataMgr.getReverseAliasMap();
for i in range(len(self.stnList)):
stnId = list(translate[self.stnList[i]])[0];
stnId = stnId.lower().replace('::','_');
stnIdList.append(stnId);
# initialize new reference frame
rf = pyRF.pyRF(refFrame).initForStnList(stnIdList);
npvTarget = rf.npvForEpoch(fyear);
T,npvT,stats = pyStk.helmert(self.npvF, npvTarget,reweightFactor=5,limit=1.75,maxIter=100);
status = self.getStatus(T, stats);
summary = self.__mkSummary(refFrame, npvT, status, stats, T, np.NAN);
if not fid == None:
fid.write('\n');
fid.write('Reference Frame:\n');
fid.write(summary+'\n')
else:
print
print 'Reference Frame:'
print summary;
if status == 'used':
self.npvF = npvT;
def stk():
tstart = time.time()
hfiles = hfileLib.Hsrc(2010, 103)
hfiles.addProjectSrc('glbl', 'glx',
'../data/osuGLOBAL/gamit/orbit/solutions')
hfiles.addProjectSrc('glbd', 'glx',
'../data/osuGLOBAL/gamit/densification/solutions')
hfiles.addProjectSrc('capp', 'glx', '../data/CAP/gamit/solutions')
hfiles.addProjectSrc('anet', 'glx', '../data/ANET/gamit/solutions')
hfiles.addProjectSrc('gnet', 'glx', '../data/GNET/gamit/solutions')
# print the number of hfiles in the system
print hfiles.size()
# resolve any conflicting names
hfiles.resolveNameConflicts()
# get the stn list for npv syncronization
stnNameList = list(set(hfiles.getStnNamesAsList()))
npvF = npvForStnList(hfiles.hfileObjList[0], stnNameList)
header = 'file status iter RMS [mm] wRMS [mm] maxResid %-out N-out N-ties N-stns Tx [mm] Ty [mm] Tz [mm] Rx [nRad] Ry[nRad] Rz [nRad]'
print header
flist = list()
for hfile in hfiles:
hfileName = hfile.getUniqueHfileName()
# print
# print
# print hfileName;
# get the npv for the hfile
npv = npvForStnList(hfile, stnNameList)
# estimate transformation
T, npvT, stats = pyStk.helmert(npv, npvF, percentile=30, reweight=5)
# if np.any(np.abs(T['t'])>0.15) \
# or np.any(np.abs(T['r'])>1e-7):
# T,npvT,stats = pyStk.helmert(npv, npvGlobal);
status = 'used'
if stats['iter'] >= stats['maxIter']:
status = 'failed'
flist.append(hfileName + ' failed to converge at ' +
str(stats['iter']) + ' iterations')
if stats['npts'] < 2:
#print hfileName,'failed b/c only has ',stats['npts'],'common tie stations';
#continue
status = 'failed'
flist.append(hfileName + ' failed b/c only has ' +
str(stats['npts']) + ' common tie stations')
if np.any(np.abs(T['t'])>0.5) \
or np.any(np.abs(T['r'])>1e-6):
status = 'failed'
flist.append(hfileName + ' failed with crazy transform')
# combine with new solution
if status == 'used':
npvF = combine(npvT.copy(), npvF.copy()).copy()
if not np.any(np.abs(T['t'])>0.5) \
and not np.any(np.abs(T['r'])>1e-6):
T['t'] = T['t'] / 1e-3
T['r'] = T['r'] / 1e-9
else:
T['t'] = np.ones_like(T['t']) * np.nan
T['r'] = np.ones_like(T['r']) * np.nan
if stats['RMS'] / 1e-3 > 100:
stats['RMS'] = np.nan
if stats['wRMS'] / 1e-3 > 100:
stats['wRMS'] = np.nan
id_status = '%s %6s %4d' % (hfileName, status, stats['iter'])
rms = '%8.2f %8.2f %11.2f' % (stats['RMS'] / 1e-3, stats['wRMS'] /
1e-3, stats['dvMax'] / 1e-3)
pout = ' %4.1f %4d %4d %4d' % (
stats['pout'], stats['nout'], stats['npts'],
np.nonzero(~np.isnan(npv))[0].size / 3)
transform = ' %10.4f %10.4f %10.4f %10.4f %10.4f %10.4f' % (
T['t'][0], T['t'][1], T['t'][2], T['r'][0], T['r'][1], T['r'][2])
print id_status, rms, pout, transform
print
for l in flist:
print l
print 'elapsed time: ', time.time() - tstart
return hfiles, stnNameList, npvF
def transformWithScaleForEpoch(self, npv, fyear):
npvTarget = self.npvForEpoch(fyear)
T, npvT, stats = pyStk.helmert(npv, npvTarget, withScale='yes')
return T, npvT, stats
def transformForEpoch(self, npv, fyear):
npvTarget = self.npvForEpoch(fyear)
T, npvT, stats = pyStk.helmert(npv, npvTarget)
return T, npvT, stats
def stk(self,**kwargs):
# get the data manager
dataMgr = self.dataMgr;
# check if there is more than 1 network
if len(dataMgr) == 1:
return;
iterCount = 0;
residualIQR = 0.025;
# something very large
previousResidualIQR = 1e10;
# init with original
previousNpvsStacked = self.npvsStacked.copy();
# flag final iteration
isFinalIter = False;
# number of coordinates = numStns *3
numCoords = np.sum(~np.isnan(self.npvs));
while iterCount < self.maxIter:
self.summary.append('iter = '+str(iterCount))
self.summary.append(self.header);
# align each column (npv) onto the npvTarget
for j in range(0,self.npvs.shape[1]):
# align the j'th column onto the target
try:
T,npvT,stats \
= pyStk.helmert(self.npvs[:,j], \
self.npvTarget, \
sigma=residualIQR \
)
# save the aligned npv
self.npvsStacked[:,j] = npvT.flatten().copy();
# save the weights associated with each network alignment
self.npvsWeights[:,j] = stats['weights'].flatten().copy();
except Exception, e:
#raise;
print dataMgr[j].name(),e;
self.npvsStacked[:,j] = np.ones_like(self.npvsStacked[:,j])*np.nan;
self.npvsWeights[:,j] = np.zeros_like(self.npvsStacked[:,j]);
# compute the residuals
self.npvsRedisuals[:,j] = self.npvTarget - self.npvsStacked[:,j];
# figure out of alignment was successful
status = self.getStatus(T, stats);
# if the network failed then zero it out via weights
if status == 'failed':
self.npvsWeights[:,j] = np.zeros_like(self.npvsStacked[:,j]);
self.npvsWeights[np.isnan(self.npvs)] = np.nan;
# record the stats and transformation parameters
self.stats.append(stats);
self.transforms.append(T);
# record the summary for this network iteration
self.summary.append(self.__mkSummary(dataMgr[j].name(), npvT, status, stats, T, np.NAN));
# update the iteration count
iterCount = iterCount + 1;
# compute the residualIQR of all the residuals
residualIQR = pyWeights.ipr(self.npvsRedisuals.flatten(), 50);
# compute difference between current and previous iteration IQRs
diffIQR = previousResidualIQR - residualIQR;
# compute the percentage of down weighted data
pout = (np.sum(self.npvsWeights < 1)/(numCoords*1.0))*100.0;
p1 = np.sum(self.npvsWeights[self.npvsWeights < 1 ] >= 0.5);
p1 = (p1/(numCoords*1.0))*100;
p2 = (np.sum(self.npvsWeights < 0.5)/(numCoords*1.0))*100;
# generate some summary stats for this iteration
self.summary.append( "pySTK: residualIQR = %6.2f [mm]" % (residualIQR/1e-3));
self.summary.append( "pySTK: decrease in sigma = %6.2f [mm]" % (diffIQR/1e-3));
self.summary.append( "pySTK: percentage of downweighted data = %4.1f, %4.1f, %4.1f" % (pout,p1,p2));
# if this is the final iteration then we're done
if isFinalIter:
break;
# check for convergence
if np.any((diffIQR) <self.sigtol):
isFinalIter = True;
# special case that current iteration actually made IQR worse
# if so roll back to previous iteration and boost the IQR sigma
if diffIQR < 0 or pout > 35:
# figure out how much to increase sigma
increment = max(0.002, abs(diffIQR), (residualIQR+abs(diffIQR))/1.5);
# restore previous iteration sigma plus fudge factor
residualIQR = previousResidualIQR + increment;
# restore previous iteration aligned networks
self.npvsStacked = previousNpvsStacked.copy();
# blab about it
self.summary.append( "pySTK: rolling back residualIQR = %6.2f [mm]" % (residualIQR/1e-3));
# recompute the target from stacked npv
self.npvTarget = nanMedianForRows(self.npvsStacked,2);
# save the stations for next round
previousResidualIQR = residualIQR;
previousNpvsStacked = self.npvsStacked.copy();
# for readability
self.summary.append("");
#print npv.shape
ts = pyStk.pyTS().initFromMatFile('../data/ts.mat');
rf = pyRF.pyRF('itrf08').initForStnList(map(str.lower,ts.stn_list));
fyear = ts.epochs[4000];
npv = ts.npvForEpoch(fyear);
npvRF = rf.npvForEpoch(fyear);
print "mean: ", stats.nanmean(npv-npvRF)
print "median: ", stats.nanmedian(npv-npvRF)
print "Aligning epoch ",fyear
T,npvT,stats = pyStk.helmert(npv, npvRF);
print
print 'iter:',stats['iter'];
print
print 'pout:',stats['pout'];
print 'nout:',stats['nout'];
print 'npts:',stats['npts'];
print
print ' RMS:',stats['RMS']/1e-3, '[mm]'
print 'wRMS:',stats['wRMS']/1e-3,'[mm]';
print
print 'max resid:',stats['dvMax']/1e-3,'[mm]'
print 'max resid indx:',stats['dvMaxIndx'][0]
print 'max resid stn:',ts.stn_list[stats['dvMaxIndx']/3]