def PrepareJHJ_EKF(self,Pa_in,rms):
self.L_JHJinv=[]
incr=1
# pylab.figure(1)
# pylab.clf()
# pylab.imshow(np.abs(self.JHJ),interpolation="nearest")
# pylab.draw()
# pylab.show(False)
# pylab.pause(0.1)
for ipol in range(self.NJacobBlocks_X):
PaPol=self.GivePaPol(Pa_in,ipol)
Pinv=ModLinAlg.invSVD(PaPol)
JHJ=self.L_JHJ[ipol]#*(1./rms**2)
JHJ+=Pinv
if self.DoReg:
JHJ+=self.LQxInv[ipol]*(self.gamma**2)
JHJinv=ModLinAlg.invSVD(JHJ)
self.L_JHJinv.append(JHJinv)
def PrepareJHJ_LM(self):
self.L_JHJinv=[]
if self.DataAllFlagged:
return
for ipol in range(self.NJacobBlocks_X):
M=self.L_JHJ[ipol]
if self.DoTikhonov:
self.LambdaTkNorm=self.LambdaTk*np.mean(np.abs(np.diag(M)))
# Lin.shape= (self.NDir,self.NJacobBlocks_X,self.NJacobBlocks_Y)
Linv=np.diag(self.Linv[:,ipol,:].ravel())
Linv*=self.LambdaTkNorm/(1.+self.LambdaLM)
M2=M+Linv
# pylab.clf()
# pylab.subplot(1,2,1)
# pylab.imshow(np.abs(M),interpolation="nearest")
# pylab.colorbar()
# pylab.subplot(1,2,2)
# pylab.imshow(np.abs(Linv),interpolation="nearest")
# pylab.colorbar()
# pylab.draw()
# pylab.show(False)
# pylab.pause(0.1)
# stop
else:
M2=M
JHJinv=ModLinAlg.invSVD(M2)
#JHJinv=ModLinAlg.invSVD(self.JHJ)
self.L_JHJinv.append(JHJinv)
def Stack_SingleTime(self, iA0, iA1):
self.DicoDATA.reload()
self.DicoGrids.reload()
DicoMSInfos = self.DicoMSInfos
#indRow = np.where((self.DicoDATA["times"]==self.times[iTime]))[0]
#ch0,ch1=self.DicoGrids["DomainEdges_Freq"][iFreq],self.DicoGrids["DomainEdges_Freq"][iFreq+1]
nrow, nch, _ = self.DicoDATA["data"].shape
ch0, ch1 = 0, nch
#print(ch0,ch1)
C0 = (self.DicoDATA["A0"] == iA0) & (self.DicoDATA["A1"] == iA1)
indRow = np.where(C0)[0]
f = self.DicoDATA["flag"][indRow, ch0:ch1, :]
d = self.DicoDATA["data"][indRow, ch0:ch1, :]
dp = self.DicoDATA["data_p"][indRow, ch0:ch1, :]
nrow, nch, _ = d.shape
weights = (self.DicoDATA["weights"][indRow, ch0:ch1]).reshape(
(nrow, nch, 1))
A0s = self.DicoDATA["A0"][indRow]
A1s = self.DicoDATA["A1"][indRow]
times = self.DicoDATA["times"][indRow]
NTimesBin = np.unique(times).size
i_TimeBin = np.int64(
np.argmin(np.abs(
times.reshape((-1, 1)) -
np.sort(np.unique(times)).reshape((1, -1))),
axis=1).reshape(-1, 1)) * np.ones((1, nch))
i_ch = np.int64((np.arange(nch).reshape(1, -1)) * np.ones((nrow, 1)))
i_A0 = A0s.reshape((-1, 1)) * np.ones((1, nch))
i_A1 = A1s.reshape((-1, 1)) * np.ones((1, nch))
i_TimeBin = np.int64(i_TimeBin).ravel()
i_ch = np.int64(i_ch).ravel()
i_A0 = np.int64(i_A0).ravel()
i_A1 = np.int64(i_A1).ravel()
na = self.na
def Give_R(din):
d0 = (din[:, :, 0] + din[:, :, -1]) / 2.
R = np.zeros((nch, NTimesBin), din.dtype)
ind0 = i_ch * NTimesBin + i_TimeBin
R.flat[:] = d0.flat[ind0]
R = d0.T
Rf = np.ones((R.shape), dtype=np.float64)
Rf[R == 0] = 0
return R, Rf
iMS = self.DicoDATA["iMS"]
f0, _ = self.Freq_minmax
dcorr = d
dcorr[f == 1] = 0
RMS = scipy.stats.median_abs_deviation((dcorr[dcorr != 0]).ravel(),
scale="normal")
df = (dcorr[dcorr != 0]).ravel()
if df.size > 100:
RMS = np.sqrt(np.abs(np.sum(df * df.conj())) / df.size)
dp[dp == 0] = 1.
R, Rf = Give_R(dcorr)
Rp, Rpf = Give_R(dp)
Rp[Rpf == 0] = 1.
#Rp/=np.abs(Rp)
R /= Rp
if np.max(np.abs(R)) == 0: return
C = np.dot(R.T.conj(), R)
Cn = np.dot(Rf.T, Rf)
Cn[Cn == 0] = 1.
C /= Cn
# import pylab
# pylab.clf()
# pylab.subplot(2,2,1)
# pylab.imshow(np.abs(C),aspect="auto",interpolation="nearest")
# pylab.colorbar()
# pylab.title("%i, %i"%(iA0,iA1))
# pylab.subplot(2,2,2)
# pylab.imshow(np.abs(R),aspect="auto",interpolation="nearest")
# pylab.colorbar()
# pylab.subplot(2,2,3)
# pylab.imshow(np.abs(Cn),aspect="auto",interpolation="nearest")
# pylab.colorbar()
# pylab.title("%i, %i"%(iA0,iA1))
# pylab.subplot(2,2,4)
# pylab.imshow(np.abs(Rf),aspect="auto",interpolation="nearest")
# pylab.colorbar()
# pylab.draw()
# pylab.show(block=False)
# pylab.pause(0.1)
# return
# ################################
Cp = np.dot(Rp.T.conj(), Rp)
Cpn = np.dot(Rpf.T, Rpf)
Cpn[Cpn == 0] = 1.
Cp /= Cpn
C /= Cp
C = np.abs(C)
invC = ModLinAlg.invSVD(C[:, :])
w = np.abs(np.sum(invC, axis=0))
WOUT = self.DicoDATA["WOUT"][indRow, ch0:ch1]
for ii, iRow in enumerate(indRow):
for ich in np.arange(ch0, ch1):
self.DicoDATA["WOUT"][iRow, ich] = w[ii]