[yt-svn] commit/yt: 3 new changesets
commits-noreply at bitbucket.org
commits-noreply at bitbucket.org
Fri Mar 14 05:25:51 PDT 2014
3 new commits in yt:
https://bitbucket.org/yt_analysis/yt/commits/538d730747c0/
Changeset: 538d730747c0
Branch: yt
User: hegan
Date: 2014-02-18 23:37:38
Summary: Checking for numerical instability in absorption spectrum fits
Affected #: 1 file
diff -r ae241d57d2f49527998ce5fc56271345fef038a8 -r 538d730747c0210045772ce80ed2a13296266895 yt/analysis_modules/absorption_spectrum/absorption_spectrum_fit.py
--- a/yt/analysis_modules/absorption_spectrum/absorption_spectrum_fit.py
+++ b/yt/analysis_modules/absorption_spectrum/absorption_spectrum_fit.py
@@ -4,10 +4,9 @@
from yt.analysis_modules.absorption_spectrum.absorption_line \
import voigt
-
def generate_total_fit(x, fluxData, orderFits, speciesDicts,
- minError=1E-5, complexLim=.999,
- fitLim=.99, minLength=3,
+ minError=None, complexLim=.995,
+ fitLim=.97, minLength=3,
maxLength=1000, splitLim=.99,
output_file=None):
@@ -90,6 +89,7 @@
fluxData[0]=1
fluxData[-1]=1
+
#Find all regions where lines/groups of lines are present
cBounds = _find_complexes(x, fluxData, fitLim=fitLim,
complexLim=complexLim, minLength=minLength,
@@ -111,6 +111,7 @@
yDatBounded=fluxData[b[1]:b[2]]
yFitBounded=yFit[b[1]:b[2]]
+
#Find init redshift
z=(xBounded[yDatBounded.argmin()]-initWl)/initWl
@@ -121,24 +122,33 @@
#Fit Using complex tools
newLinesP,flag=_complex_fit(xBounded,yDatBounded,yFitBounded,
- z,fitLim,minError*(b[2]-b[1]),speciesDict)
+ z,fitLim,minError,speciesDict)
+
+ #If flagged as a bad fit, species is lyman alpha,
+ # and it may be a saturated line, use special tools
+ if flag and species=='lya' and min(yDatBounded)<.1:
+ newLinesP=_large_flag_fit(xBounded,yDatBounded,
+ yFitBounded,z,speciesDict,
+ minSize,minError)
+
+ if na.size(newLinesP)> 0:
+
+ #Check for EXPLOOOOSIIONNNSSS
+ newLinesP = _check_numerical_instability(x, newLinesP, speciesDict,b)
+
#Check existence of partner lines if applicable
if len(speciesDict['wavelength']) != 1:
newLinesP = _remove_unaccepted_partners(newLinesP, x, fluxData,
- b, minError*(b[2]-b[1]),
- x0, xRes, speciesDict)
+ b, minError, x0, xRes, speciesDict)
- #If flagged as a bad fit, species is lyman alpha,
- # and it may be a saturated line, use special tools
- if flag and species=='lya' and min(yDatBounded)<.1:
- newLinesP=_large_flag_fit(xBounded,yDatBounded,
- yFitBounded,z,speciesDict,
- minSize,minError*(b[2]-b[1]))
+
+
#Adjust total current fit
yFit=yFit*_gen_flux_lines(x,newLinesP,speciesDict)
+
#Add new group to all fitted lines
if na.size(newLinesP)>0:
speciesLines['N']=na.append(speciesLines['N'],newLinesP[:,0])
@@ -149,6 +159,7 @@
allSpeciesLines[species]=speciesLines
+
if output_file:
_output_fit(allSpeciesLines, output_file)
@@ -205,10 +216,12 @@
#Setup initial line guesses
if initP==None: #Regular fit
initP = [0,0,0]
- if min(yDat)<.5: #Large lines get larger initial guess
- initP[0] = 10**16
+ if min(yDat)<.01: #Large lines get larger initial guess
+ initP[0] = speciesDict['init_N']*10**2
+ elif min(yDat)<.5:
+ initP[0] = speciesDict['init_N']*10**1
elif min(yDat)>.9: #Small lines get smaller initial guess
- initP[0] = 10**12.5
+ initP[0] = speciesDict['init_N']*10**-1
else:
initP[0] = speciesDict['init_N']
initP[1] = speciesDict['init_b']
@@ -225,9 +238,16 @@
return [],False
#Values to proceed through first run
- errSq,prevErrSq=1,1000
+ errSq,prevErrSq,prevLinesP=1,10*len(x),[]
+ if errBound == None:
+ errBound = len(yDat)*(max(1-yDat)*1E-2)**2
+ else:
+ errBound = errBound*len(yDat)
+
+ flag = False
while True:
+
#Initial parameter guess from joining parameters from all lines
# in lines into a single array
initP = linesP.flatten()
@@ -237,6 +257,7 @@
args=(x,yDat,yFit,speciesDict),
epsfcn=1E-10,maxfev=1000)
+
#Set results of optimization
linesP = na.reshape(fitP,(-1,3))
@@ -247,17 +268,23 @@
#Sum to get idea of goodness of fit
errSq=sum(dif**2)
+ if any(linesP[:,1]==speciesDict['init_b']):
+ # linesP = prevLinesP
+
+ flag = True
+ break
+
#If good enough, break
- if errSq < errBound:
+ if errSq < errBound:
break
#If last fit was worse, reject the last line and revert to last fit
- if errSq > prevErrSq*10:
+ if errSq > prevErrSq*10 :
#If its still pretty damn bad, cut losses and try flag fit tools
if prevErrSq >1E2*errBound and speciesDict['name']=='HI lya':
return [],True
else:
- yNewFit=_gen_flux_lines(x,prevLinesP,speciesDict)
+ linesP = prevLinesP
break
#If too many lines
@@ -266,21 +293,26 @@
if errSq >1E2*errBound and speciesDict['name']=='HI lya':
return [],True
else:
- break
+ flag = True
+ break
#Store previous data in case reject next fit
prevErrSq = errSq
prevLinesP = linesP
-
#Set up initial condition for new line
newP = [0,0,0]
- if min(dif)<.1:
- newP[0]=10**12
- elif min(dif)>.9:
- newP[0]=10**16
+
+ yAdjusted = 1+yFit*yNewFit-yDat
+
+ if min(yAdjusted)<.01: #Large lines get larger initial guess
+ newP[0] = speciesDict['init_N']*10**2
+ elif min(yAdjusted)<.5:
+ newP[0] = speciesDict['init_N']*10**1
+ elif min(yAdjusted)>.9: #Small lines get smaller initial guess
+ newP[0] = speciesDict['init_N']*10**-1
else:
- newP[0]=10**14
+ newP[0] = speciesDict['init_N']
newP[1] = speciesDict['init_b']
newP[2]=(x[dif.argmax()]-wl0)/wl0
linesP=na.append(linesP,[newP],axis=0)
@@ -290,12 +322,12 @@
# acceptable range, as given in dict ref
remove=[]
for i,p in enumerate(linesP):
- check=_check_params(na.array([p]),speciesDict)
+ check=_check_params(na.array([p]),speciesDict,x)
if check:
remove.append(i)
linesP = na.delete(linesP,remove,axis=0)
- return linesP,False
+ return linesP,flag
def _large_flag_fit(x, yDat, yFit, initz, speciesDict, minSize, errBound):
"""
@@ -489,6 +521,9 @@
#List of lines to remove
removeLines=[]
+ #Set error
+
+
#Iterate through all sets of line parameters
for i,p in enumerate(linesP):
@@ -501,16 +536,23 @@
lb = _get_bounds(p[2],b,wl,x0,xRes)
xb,yb=x[lb[0]:lb[1]],y[lb[0]:lb[1]]
+ if errBound == None:
+ errBound = 10*len(yb)*(max(1-yb)*1E-2)**2
+ else:
+ errBound = 10*errBound*len(yb)
+
#Generate a fit and find the difference to data
yFitb=_gen_flux_lines(xb,na.array([p]),speciesDict)
dif =yb-yFitb
+
+
#Only counts as an error if line is too big ---------------<
dif = [k for k in dif if k>0]
err = sum(dif)
#If the fit is too bad then add the line to list of removed lines
- if err > errBound*1E2:
+ if err > errBound:
removeLines.append(i)
break
@@ -640,21 +682,13 @@
#Check if the region needs to be divided
if b[2]-b[1]>maxLength:
- #Find the minimum absorption in the middle two quartiles of
- # the large complex
- q=(b[2]-b[1])/4
- cut = yDat[b[1]+q:b[2]-q].argmax()+b[1]+q
+ split = _split_region(yDat,b,splitLim)
- #Only break it up if the minimum absorption is actually low enough
- if yDat[cut]>splitLim:
-
- #Get the new two peaks
- b1Peak = yDat[b[1]:cut].argmin()+b[1]
- b2Peak = yDat[cut:b[2]].argmin()+cut
+ if split:
#add the two regions separately
- cBounds.insert(i+1,[b1Peak,b[1],cut])
- cBounds.insert(i+2,[b2Peak,cut,b[2]])
+ cBounds.insert(i+1,split[0])
+ cBounds.insert(i+2,split[1])
#Remove the original region
cBounds.pop(i)
@@ -663,7 +697,33 @@
return cBounds
-def _gen_flux_lines(x, linesP, speciesDict):
+
+def _split_region(yDat,b,splitLim):
+ #Find the minimum absorption in the middle two quartiles of
+ # the large complex
+
+ q=(b[2]-b[1])/4
+ cut = yDat[b[1]+q:b[2]-q].argmax()+b[1]+q
+
+ #Only break it up if the minimum absorption is actually low enough
+ if yDat[cut]>splitLim:
+
+ #Get the new two peaks
+ b1Peak = yDat[b[1]:cut].argmin()+b[1]
+ b2Peak = yDat[cut:b[2]].argmin()+cut
+
+ region_1 = [b1Peak,b[1],cut]
+ region_2 = [b2Peak,cut,b[2]]
+
+ return [region_1,region_2]
+
+ else:
+
+ return []
+
+
+
+def _gen_flux_lines(x, linesP, speciesDict,firstLine=False):
"""
Calculates the normalized flux for a region of wavelength space
generated by a set of absorption lines.
@@ -692,6 +752,9 @@
g=speciesDict['Gamma'][i]
wl=speciesDict['wavelength'][i]
y = y+ _gen_tau(x,p,f,g,wl)
+ if firstLine:
+ break
+
flux = na.exp(-y)
return flux
@@ -744,21 +807,25 @@
the difference between the fit generated by the parameters
given in pTotal multiplied by the previous fit and the desired
flux profile, w/ first index modified appropriately for bad
- parameter choices
+ parameter choices and additional penalty for fitting with a lower
+ flux than observed.
"""
pTotal.shape = (-1,3)
yNewFit = _gen_flux_lines(x,pTotal,speciesDict)
error = yDat-yFit*yNewFit
- error[0] = _check_params(pTotal,speciesDict)
+ error_plus = (yDat-yFit*yNewFit).clip(min=0)
+
+ error = error+error_plus
+ error[0] = _check_params(pTotal,speciesDict,x)
return error
-def _check_params(p, speciesDict):
+def _check_params(p, speciesDict,xb):
"""
Check to see if any of the parameters in p fall outside the range
- given in speciesDict.
+ given in speciesDict or on the boundaries
Parameters
----------
@@ -767,6 +834,8 @@
speciesDict : dictionary
dictionary with properties giving the max and min
values appropriate for each parameter N,b, and z.
+ xb : (N) ndarray
+ wavelength array [nm]
Returns
-------
@@ -774,16 +843,137 @@
0 if all values are fine
999 if any values fall outside acceptable range
"""
+
+ minz = (xb[0])/speciesDict['wavelength'][0]-1
+ maxz = (xb[-1])/speciesDict['wavelength'][0]-1
+
check = 0
- if any(p[:,0] > speciesDict['maxN']) or\
- any(p[:,0] < speciesDict['minN']) or\
- any(p[:,1] > speciesDict['maxb']) or\
- any(p[:,1] < speciesDict['minb']) or\
- any(p[:,2] > speciesDict['maxz']) or\
- any(p[:,2] < speciesDict['minz']):
+ if any(p[:,0] >= speciesDict['maxN']) or\
+ any(p[:,0] <= speciesDict['minN']) or\
+ any(p[:,1] >= speciesDict['maxb']) or\
+ any(p[:,1] <= speciesDict['minb']) or\
+ any(p[:,2] >= maxz) or\
+ any(p[:,2] <= minz):
check = 999
+
return check
+def _check_optimization_init(p,speciesDict,initz,xb,yDat,yFit,minSize,errorBound):
+
+ """
+ Check to see if any of the parameters in p are the
+ same as initial paramters and if so, attempt to
+ split the region and refit it.
+
+ Parameters
+ ----------
+ p : (3,) ndarray
+ array with form [[N1, b1, z1], ...]
+ speciesDict : dictionary
+ dictionary with properties giving the max and min
+ values appropriate for each parameter N,b, and z.
+ x : (N) ndarray
+ wavelength array [nm]
+ """
+
+ # Check if anything is a default parameter
+ if any(p[:,0] == speciesDict['init_N']) or\
+ any(p[:,0] == speciesDict['init_N']*10) or\
+ any(p[:,0] == speciesDict['init_N']*100) or\
+ any(p[:,0] == speciesDict['init_N']*.1) or\
+ any(p[:,1] == speciesDict['init_b']) or\
+ any(p[:,1] == speciesDict['maxb']):
+
+ # These are the initial bounds
+ init_bounds = [yDat.argmin(),0,len(xb)-1]
+
+ # Gratitutous limit for splitting region
+ newSplitLim = 1 - (1-min(yDat))*.5
+
+ # Attempt to split region
+ split = _split_region(yDat,init_bounds,newSplitLim)
+
+ # If we can't split it, just reject it. Its unphysical
+ # to just keep the default parameters and we're out of
+ # options at this point
+ if not split:
+ return []
+
+ # Else set up the bounds for each region and fit separately
+ b1,b2 = split[0][2], split[1][1]
+
+ p1,flag = _complex_fit(xb[:b1], yDat[:b1], yFit[:b1],
+ initz, minSize, errorBound, speciesDict)
+
+ p2,flag = _complex_fit(xb[b2:], yDat[b2:], yFit[b2:],
+ initz, minSize, errorBound, speciesDict)
+
+ # Make the final line parameters. Its annoying because
+ # one or both regions may have fit to nothing
+ if na.size(p1)> 0 and na.size(p2)>0:
+ p = na.r_[p1,p2]
+ elif na.size(p1) > 0:
+ p = p1
+ else:
+ p = p2
+
+ return p
+
+
+def _check_numerical_instability(x, p, speciesDict,b):
+
+ """
+ Check to see if any of the parameters in p are causing
+ unstable numerical effects outside the region of fit
+
+ Parameters
+ ----------
+ p : (3,) ndarray
+ array with form [[N1, b1, z1], ...]
+ speciesDict : dictionary
+ dictionary with properties giving the max and min
+ values appropriate for each parameter N,b, and z.
+ x : (N) ndarray
+ wavelength array [nm]
+ b : (3) list
+ list of integers indicating bounds of region fit in x
+ """
+
+ remove_lines = []
+
+
+ for i,line in enumerate(p):
+
+ # First to check if the line is at risk for instability
+ if line[1]<5 or line[0] < 1E12:
+
+
+ # get all flux that isn't part of fit plus a little wiggle room
+ # max and min to prevent boundary errors
+
+ flux = _gen_flux_lines(x,[line],speciesDict,firstLine=True)
+ flux = na.r_[flux[:max(b[1]-10,0)], flux[min(b[2]+10,len(x)):]]
+
+ #Find regions that are absorbing outside the region we fit
+ flux_dif = 1 - flux
+ absorbing_coefficient = max(abs(flux_dif))
+
+
+ #Really there shouldn't be any absorption outside
+ #the region we fit, but we'll give some leeway.
+ #for high resolution spectra the tiny bits on the edges
+ #can give a non negligible amount of flux. Plus the errors
+ #we are looking for are HUGE.
+ if absorbing_coefficient > .1:
+
+ # we just set it to no fit because we've tried
+ # everything else at this point. this region just sucks :(
+ remove_lines.append(i)
+
+ if remove_lines:
+ p = na.delete(p, remove_lines, axis=0)
+
+ return p
def _output_fit(lineDic, file_name = 'spectrum_fit.h5'):
"""
@@ -815,4 +1005,5 @@
f.create_dataset("{0}/z".format(ion),data=params['z'])
f.create_dataset("{0}/complex".format(ion),data=params['group#'])
print 'Writing spectrum fit to {0}'.format(file_name)
+ f.close()
https://bitbucket.org/yt_analysis/yt/commits/89ffc3dc0dbc/
Changeset: 89ffc3dc0dbc
Branch: yt
User: hegan
Date: 2014-02-18 23:53:40
Summary: minError = None, not ready for public yet
Affected #: 1 file
diff -r 538d730747c0210045772ce80ed2a13296266895 -r 89ffc3dc0dbc1d4c800e384b7137aefd47de5fb8 yt/analysis_modules/absorption_spectrum/absorption_spectrum_fit.py
--- a/yt/analysis_modules/absorption_spectrum/absorption_spectrum_fit.py
+++ b/yt/analysis_modules/absorption_spectrum/absorption_spectrum_fit.py
@@ -5,7 +5,7 @@
import voigt
def generate_total_fit(x, fluxData, orderFits, speciesDicts,
- minError=None, complexLim=.995,
+ minError=1E-4, complexLim=.995,
fitLim=.97, minLength=3,
maxLength=1000, splitLim=.99,
output_file=None):
https://bitbucket.org/yt_analysis/yt/commits/837706704f10/
Changeset: 837706704f10
Branch: yt
User: MatthewTurk
Date: 2014-03-14 13:25:42
Summary: Merged in hegan/yt (pull request #709)
Checking for numerical instability in absorption spectrum fits
Affected #: 1 file
diff -r 61e6b84f875cc8fcf25b5b1e67ddd501a19daf68 -r 837706704f10f8653e3a4f9b9367b104b047424d yt/analysis_modules/absorption_spectrum/absorption_spectrum_fit.py
--- a/yt/analysis_modules/absorption_spectrum/absorption_spectrum_fit.py
+++ b/yt/analysis_modules/absorption_spectrum/absorption_spectrum_fit.py
@@ -4,10 +4,9 @@
from yt.analysis_modules.absorption_spectrum.absorption_line \
import voigt
-
def generate_total_fit(x, fluxData, orderFits, speciesDicts,
- minError=1E-5, complexLim=.999,
- fitLim=.99, minLength=3,
+ minError=1E-4, complexLim=.995,
+ fitLim=.97, minLength=3,
maxLength=1000, splitLim=.99,
output_file=None):
@@ -90,6 +89,7 @@
fluxData[0]=1
fluxData[-1]=1
+
#Find all regions where lines/groups of lines are present
cBounds = _find_complexes(x, fluxData, fitLim=fitLim,
complexLim=complexLim, minLength=minLength,
@@ -111,6 +111,7 @@
yDatBounded=fluxData[b[1]:b[2]]
yFitBounded=yFit[b[1]:b[2]]
+
#Find init redshift
z=(xBounded[yDatBounded.argmin()]-initWl)/initWl
@@ -121,24 +122,33 @@
#Fit Using complex tools
newLinesP,flag=_complex_fit(xBounded,yDatBounded,yFitBounded,
- z,fitLim,minError*(b[2]-b[1]),speciesDict)
+ z,fitLim,minError,speciesDict)
+
+ #If flagged as a bad fit, species is lyman alpha,
+ # and it may be a saturated line, use special tools
+ if flag and species=='lya' and min(yDatBounded)<.1:
+ newLinesP=_large_flag_fit(xBounded,yDatBounded,
+ yFitBounded,z,speciesDict,
+ minSize,minError)
+
+ if na.size(newLinesP)> 0:
+
+ #Check for EXPLOOOOSIIONNNSSS
+ newLinesP = _check_numerical_instability(x, newLinesP, speciesDict,b)
+
#Check existence of partner lines if applicable
if len(speciesDict['wavelength']) != 1:
newLinesP = _remove_unaccepted_partners(newLinesP, x, fluxData,
- b, minError*(b[2]-b[1]),
- x0, xRes, speciesDict)
+ b, minError, x0, xRes, speciesDict)
- #If flagged as a bad fit, species is lyman alpha,
- # and it may be a saturated line, use special tools
- if flag and species=='lya' and min(yDatBounded)<.1:
- newLinesP=_large_flag_fit(xBounded,yDatBounded,
- yFitBounded,z,speciesDict,
- minSize,minError*(b[2]-b[1]))
+
+
#Adjust total current fit
yFit=yFit*_gen_flux_lines(x,newLinesP,speciesDict)
+
#Add new group to all fitted lines
if na.size(newLinesP)>0:
speciesLines['N']=na.append(speciesLines['N'],newLinesP[:,0])
@@ -149,6 +159,7 @@
allSpeciesLines[species]=speciesLines
+
if output_file:
_output_fit(allSpeciesLines, output_file)
@@ -205,10 +216,12 @@
#Setup initial line guesses
if initP==None: #Regular fit
initP = [0,0,0]
- if min(yDat)<.5: #Large lines get larger initial guess
- initP[0] = 10**16
+ if min(yDat)<.01: #Large lines get larger initial guess
+ initP[0] = speciesDict['init_N']*10**2
+ elif min(yDat)<.5:
+ initP[0] = speciesDict['init_N']*10**1
elif min(yDat)>.9: #Small lines get smaller initial guess
- initP[0] = 10**12.5
+ initP[0] = speciesDict['init_N']*10**-1
else:
initP[0] = speciesDict['init_N']
initP[1] = speciesDict['init_b']
@@ -225,9 +238,16 @@
return [],False
#Values to proceed through first run
- errSq,prevErrSq=1,1000
+ errSq,prevErrSq,prevLinesP=1,10*len(x),[]
+ if errBound == None:
+ errBound = len(yDat)*(max(1-yDat)*1E-2)**2
+ else:
+ errBound = errBound*len(yDat)
+
+ flag = False
while True:
+
#Initial parameter guess from joining parameters from all lines
# in lines into a single array
initP = linesP.flatten()
@@ -237,6 +257,7 @@
args=(x,yDat,yFit,speciesDict),
epsfcn=1E-10,maxfev=1000)
+
#Set results of optimization
linesP = na.reshape(fitP,(-1,3))
@@ -247,17 +268,23 @@
#Sum to get idea of goodness of fit
errSq=sum(dif**2)
+ if any(linesP[:,1]==speciesDict['init_b']):
+ # linesP = prevLinesP
+
+ flag = True
+ break
+
#If good enough, break
- if errSq < errBound:
+ if errSq < errBound:
break
#If last fit was worse, reject the last line and revert to last fit
- if errSq > prevErrSq*10:
+ if errSq > prevErrSq*10 :
#If its still pretty damn bad, cut losses and try flag fit tools
if prevErrSq >1E2*errBound and speciesDict['name']=='HI lya':
return [],True
else:
- yNewFit=_gen_flux_lines(x,prevLinesP,speciesDict)
+ linesP = prevLinesP
break
#If too many lines
@@ -266,21 +293,26 @@
if errSq >1E2*errBound and speciesDict['name']=='HI lya':
return [],True
else:
- break
+ flag = True
+ break
#Store previous data in case reject next fit
prevErrSq = errSq
prevLinesP = linesP
-
#Set up initial condition for new line
newP = [0,0,0]
- if min(dif)<.1:
- newP[0]=10**12
- elif min(dif)>.9:
- newP[0]=10**16
+
+ yAdjusted = 1+yFit*yNewFit-yDat
+
+ if min(yAdjusted)<.01: #Large lines get larger initial guess
+ newP[0] = speciesDict['init_N']*10**2
+ elif min(yAdjusted)<.5:
+ newP[0] = speciesDict['init_N']*10**1
+ elif min(yAdjusted)>.9: #Small lines get smaller initial guess
+ newP[0] = speciesDict['init_N']*10**-1
else:
- newP[0]=10**14
+ newP[0] = speciesDict['init_N']
newP[1] = speciesDict['init_b']
newP[2]=(x[dif.argmax()]-wl0)/wl0
linesP=na.append(linesP,[newP],axis=0)
@@ -290,12 +322,12 @@
# acceptable range, as given in dict ref
remove=[]
for i,p in enumerate(linesP):
- check=_check_params(na.array([p]),speciesDict)
+ check=_check_params(na.array([p]),speciesDict,x)
if check:
remove.append(i)
linesP = na.delete(linesP,remove,axis=0)
- return linesP,False
+ return linesP,flag
def _large_flag_fit(x, yDat, yFit, initz, speciesDict, minSize, errBound):
"""
@@ -489,6 +521,9 @@
#List of lines to remove
removeLines=[]
+ #Set error
+
+
#Iterate through all sets of line parameters
for i,p in enumerate(linesP):
@@ -501,16 +536,23 @@
lb = _get_bounds(p[2],b,wl,x0,xRes)
xb,yb=x[lb[0]:lb[1]],y[lb[0]:lb[1]]
+ if errBound == None:
+ errBound = 10*len(yb)*(max(1-yb)*1E-2)**2
+ else:
+ errBound = 10*errBound*len(yb)
+
#Generate a fit and find the difference to data
yFitb=_gen_flux_lines(xb,na.array([p]),speciesDict)
dif =yb-yFitb
+
+
#Only counts as an error if line is too big ---------------<
dif = [k for k in dif if k>0]
err = sum(dif)
#If the fit is too bad then add the line to list of removed lines
- if err > errBound*1E2:
+ if err > errBound:
removeLines.append(i)
break
@@ -640,21 +682,13 @@
#Check if the region needs to be divided
if b[2]-b[1]>maxLength:
- #Find the minimum absorption in the middle two quartiles of
- # the large complex
- q=(b[2]-b[1])/4
- cut = yDat[b[1]+q:b[2]-q].argmax()+b[1]+q
+ split = _split_region(yDat,b,splitLim)
- #Only break it up if the minimum absorption is actually low enough
- if yDat[cut]>splitLim:
-
- #Get the new two peaks
- b1Peak = yDat[b[1]:cut].argmin()+b[1]
- b2Peak = yDat[cut:b[2]].argmin()+cut
+ if split:
#add the two regions separately
- cBounds.insert(i+1,[b1Peak,b[1],cut])
- cBounds.insert(i+2,[b2Peak,cut,b[2]])
+ cBounds.insert(i+1,split[0])
+ cBounds.insert(i+2,split[1])
#Remove the original region
cBounds.pop(i)
@@ -663,7 +697,33 @@
return cBounds
-def _gen_flux_lines(x, linesP, speciesDict):
+
+def _split_region(yDat,b,splitLim):
+ #Find the minimum absorption in the middle two quartiles of
+ # the large complex
+
+ q=(b[2]-b[1])/4
+ cut = yDat[b[1]+q:b[2]-q].argmax()+b[1]+q
+
+ #Only break it up if the minimum absorption is actually low enough
+ if yDat[cut]>splitLim:
+
+ #Get the new two peaks
+ b1Peak = yDat[b[1]:cut].argmin()+b[1]
+ b2Peak = yDat[cut:b[2]].argmin()+cut
+
+ region_1 = [b1Peak,b[1],cut]
+ region_2 = [b2Peak,cut,b[2]]
+
+ return [region_1,region_2]
+
+ else:
+
+ return []
+
+
+
+def _gen_flux_lines(x, linesP, speciesDict,firstLine=False):
"""
Calculates the normalized flux for a region of wavelength space
generated by a set of absorption lines.
@@ -692,6 +752,9 @@
g=speciesDict['Gamma'][i]
wl=speciesDict['wavelength'][i]
y = y+ _gen_tau(x,p,f,g,wl)
+ if firstLine:
+ break
+
flux = na.exp(-y)
return flux
@@ -744,21 +807,25 @@
the difference between the fit generated by the parameters
given in pTotal multiplied by the previous fit and the desired
flux profile, w/ first index modified appropriately for bad
- parameter choices
+ parameter choices and additional penalty for fitting with a lower
+ flux than observed.
"""
pTotal.shape = (-1,3)
yNewFit = _gen_flux_lines(x,pTotal,speciesDict)
error = yDat-yFit*yNewFit
- error[0] = _check_params(pTotal,speciesDict)
+ error_plus = (yDat-yFit*yNewFit).clip(min=0)
+
+ error = error+error_plus
+ error[0] = _check_params(pTotal,speciesDict,x)
return error
-def _check_params(p, speciesDict):
+def _check_params(p, speciesDict,xb):
"""
Check to see if any of the parameters in p fall outside the range
- given in speciesDict.
+ given in speciesDict or on the boundaries
Parameters
----------
@@ -767,6 +834,8 @@
speciesDict : dictionary
dictionary with properties giving the max and min
values appropriate for each parameter N,b, and z.
+ xb : (N) ndarray
+ wavelength array [nm]
Returns
-------
@@ -774,16 +843,137 @@
0 if all values are fine
999 if any values fall outside acceptable range
"""
+
+ minz = (xb[0])/speciesDict['wavelength'][0]-1
+ maxz = (xb[-1])/speciesDict['wavelength'][0]-1
+
check = 0
- if any(p[:,0] > speciesDict['maxN']) or\
- any(p[:,0] < speciesDict['minN']) or\
- any(p[:,1] > speciesDict['maxb']) or\
- any(p[:,1] < speciesDict['minb']) or\
- any(p[:,2] > speciesDict['maxz']) or\
- any(p[:,2] < speciesDict['minz']):
+ if any(p[:,0] >= speciesDict['maxN']) or\
+ any(p[:,0] <= speciesDict['minN']) or\
+ any(p[:,1] >= speciesDict['maxb']) or\
+ any(p[:,1] <= speciesDict['minb']) or\
+ any(p[:,2] >= maxz) or\
+ any(p[:,2] <= minz):
check = 999
+
return check
+def _check_optimization_init(p,speciesDict,initz,xb,yDat,yFit,minSize,errorBound):
+
+ """
+ Check to see if any of the parameters in p are the
+ same as initial paramters and if so, attempt to
+ split the region and refit it.
+
+ Parameters
+ ----------
+ p : (3,) ndarray
+ array with form [[N1, b1, z1], ...]
+ speciesDict : dictionary
+ dictionary with properties giving the max and min
+ values appropriate for each parameter N,b, and z.
+ x : (N) ndarray
+ wavelength array [nm]
+ """
+
+ # Check if anything is a default parameter
+ if any(p[:,0] == speciesDict['init_N']) or\
+ any(p[:,0] == speciesDict['init_N']*10) or\
+ any(p[:,0] == speciesDict['init_N']*100) or\
+ any(p[:,0] == speciesDict['init_N']*.1) or\
+ any(p[:,1] == speciesDict['init_b']) or\
+ any(p[:,1] == speciesDict['maxb']):
+
+ # These are the initial bounds
+ init_bounds = [yDat.argmin(),0,len(xb)-1]
+
+ # Gratitutous limit for splitting region
+ newSplitLim = 1 - (1-min(yDat))*.5
+
+ # Attempt to split region
+ split = _split_region(yDat,init_bounds,newSplitLim)
+
+ # If we can't split it, just reject it. Its unphysical
+ # to just keep the default parameters and we're out of
+ # options at this point
+ if not split:
+ return []
+
+ # Else set up the bounds for each region and fit separately
+ b1,b2 = split[0][2], split[1][1]
+
+ p1,flag = _complex_fit(xb[:b1], yDat[:b1], yFit[:b1],
+ initz, minSize, errorBound, speciesDict)
+
+ p2,flag = _complex_fit(xb[b2:], yDat[b2:], yFit[b2:],
+ initz, minSize, errorBound, speciesDict)
+
+ # Make the final line parameters. Its annoying because
+ # one or both regions may have fit to nothing
+ if na.size(p1)> 0 and na.size(p2)>0:
+ p = na.r_[p1,p2]
+ elif na.size(p1) > 0:
+ p = p1
+ else:
+ p = p2
+
+ return p
+
+
+def _check_numerical_instability(x, p, speciesDict,b):
+
+ """
+ Check to see if any of the parameters in p are causing
+ unstable numerical effects outside the region of fit
+
+ Parameters
+ ----------
+ p : (3,) ndarray
+ array with form [[N1, b1, z1], ...]
+ speciesDict : dictionary
+ dictionary with properties giving the max and min
+ values appropriate for each parameter N,b, and z.
+ x : (N) ndarray
+ wavelength array [nm]
+ b : (3) list
+ list of integers indicating bounds of region fit in x
+ """
+
+ remove_lines = []
+
+
+ for i,line in enumerate(p):
+
+ # First to check if the line is at risk for instability
+ if line[1]<5 or line[0] < 1E12:
+
+
+ # get all flux that isn't part of fit plus a little wiggle room
+ # max and min to prevent boundary errors
+
+ flux = _gen_flux_lines(x,[line],speciesDict,firstLine=True)
+ flux = na.r_[flux[:max(b[1]-10,0)], flux[min(b[2]+10,len(x)):]]
+
+ #Find regions that are absorbing outside the region we fit
+ flux_dif = 1 - flux
+ absorbing_coefficient = max(abs(flux_dif))
+
+
+ #Really there shouldn't be any absorption outside
+ #the region we fit, but we'll give some leeway.
+ #for high resolution spectra the tiny bits on the edges
+ #can give a non negligible amount of flux. Plus the errors
+ #we are looking for are HUGE.
+ if absorbing_coefficient > .1:
+
+ # we just set it to no fit because we've tried
+ # everything else at this point. this region just sucks :(
+ remove_lines.append(i)
+
+ if remove_lines:
+ p = na.delete(p, remove_lines, axis=0)
+
+ return p
def _output_fit(lineDic, file_name = 'spectrum_fit.h5'):
"""
@@ -815,4 +1005,5 @@
f.create_dataset("{0}/z".format(ion),data=params['z'])
f.create_dataset("{0}/complex".format(ion),data=params['group#'])
print 'Writing spectrum fit to {0}'.format(file_name)
+ f.close()
Repository URL: https://bitbucket.org/yt_analysis/yt/
--
This is a commit notification from bitbucket.org. You are receiving
this because you have the service enabled, addressing the recipient of
this email.
More information about the yt-svn
mailing list