dispersiveXanes/xanes_analyzeRun.py

361 lines
13 KiB
Python
Raw Permalink Normal View History

2016-05-12 15:02:42 +02:00
import os
import sys
import numpy as np
np.warnings.simplefilter('ignore')
import time
import matplotlib.pyplot as plt
import h5py
2016-11-23 14:05:38 +01:00
import collections
2016-05-12 15:02:42 +02:00
import re
from x3py import x3py
import dispersiveXanes_alignment as alignment
2016-05-12 15:02:42 +02:00
import mcutils as mc
cmap = plt.cm.viridis if hasattr(plt.cm,"viridis") else plt.cm.gray
kw_2dplot = dict(
2016-05-12 15:02:42 +02:00
interpolation = "none",
aspect = "auto",
cmap = cmap
2016-05-12 15:02:42 +02:00
)
g_exp = "mecl3616"
g_exp = "xppl3716"
g_bml = g_exp[:3]
x3py.config.updateBeamline(g_bml)
basedir = os.path.dirname(__file__)
g_folder_init = basedir + "/" +g_exp+"_init_pars/"
g_folder_out = basedir + "/" +g_exp+"_output/"
2016-05-12 15:02:42 +02:00
g_folder_data = "/reg/d/psdm/"+g_bml+"/"+ g_exp +"/hdf5/"
2016-05-12 15:02:42 +02:00
import socket
hostname = socket.gethostname()
if hostname == "x1":
g_folder_data = "/home/marco/temp"
if hostname == "apcluster0":
g_folder_data = "/data/marcoc/singleShotXanes/"+ g_exp +"/hdf5/"
2016-05-12 15:02:42 +02:00
# set defaults based on experiment
if g_bml == "xpp":
g_roi_height = 200
g_swapx = False
g_swapy = False
else:
g_roi_height = 100
g_swapx = True
g_swapy = False
print("Working on experiment",g_exp,"(beamline %s)"%g_bml)
print(" folder data →",g_folder_data)
print(" folder init_pars →",g_folder_init)
print(" folder outout →",g_folder_out)
#g_folder = "/reg/d/psdm/xpp/xppl3716/ftc/hdf5/"
def readDataset(fnameOrRun=7,
force=False,
doBkgSub=False):
if isinstance(fnameOrRun,str) and (fnameOrRun[-3:]=="npz"):
d = x3py.toolsVarious.DropObject()
temp = np.load(fnameOrRun)
spec1 = temp["spec1"]
spec2 = temp["spec2"]
nS = spec1.shape[0]
d.spec1 = x3py.toolsDetectors.wrapArray("spec1",spec1,time=np.arange(nS))
d.spec2 = x3py.toolsDetectors.wrapArray("spec2",spec2,time=np.arange(nS))
else:
if isinstance(fnameOrRun,int):
fnameOrRun=g_folder_data+"/"+g_exp+"-r%04d.h5" % fnameOrRun
2016-11-23 14:05:38 +01:00
d = x3py.Dataset(fnameOrRun,detectors=["opal0","opal1","fee_spec","opal2","ebeam"])
2016-05-12 15:02:42 +02:00
if g_bml == "xpp":
d.spec1 = d.opal0
d.spec2 = d.opal1
else:
d.spec1 = d.fee_spec
d.spec2 = d.opal2
if not hasattr(d,"scan"):
d.scan = x3py.toolsVarious.DropObject()
d.scan.scanmotor0_values = [0,]
return d
def getCenter(img,axis=0,threshold=0.05):
img = img.copy()
img[img<img.max()*threshold] = 0
if axis == 1: img=img.T
p = img.mean(1)
x = np.arange(img.shape[0])
return int(np.sum(x*p)/np.sum(p))
def showShots(im1,im2):
nS = im1.shape[0]
fig,ax = plt.subplots(2,nS,sharex=True,sharey=True)
if im1.ndim == 3:
for a,i1,i2 in zip(ax.T,im1,im2):
a[0].imshow(i1.T,**kw_2dplot)
a[1].imshow(i2.T,**kw_2dplot)
else:
for a,p1,p2 in zip(ax.T,im1,im2):
a[0].plot(p1)
a[1].plot(p2)
def sliceToIndices(shot_slice,nShots):
return list(range(*shot_slice.indices(nShots)))
2016-05-12 15:02:42 +02:00
class AnalyzeRun(object):
def __init__(self,run,initAlign="auto",swapx=g_swapx,swapy=g_swapy):
""" swapx → swap x axis of first spectrometer
swapy swap y axis of first spectrometer
initAlign: could be:
1. None if you want default transformation parameters
2. a dict if you want to overwrite certain parameters of the default ones
3. an integer (to look for xppl3716_init_pars/run????_transform.npy)
4. a file name (that has been previosly saved with r.saveTransform(fname)
2016-05-12 15:02:42 +02:00
"""
2016-11-23 14:05:38 +01:00
self.data = readDataset(run)
self.scanpos = self.data.scan.scanmotor0_values
self.nCalib = self.data.spec1.nCalib
self.nShotsPerCalib = self.data.spec1.lens
2016-05-12 15:02:42 +02:00
if isinstance(run,str):
run = int( re.search("\d{3,4}",run).group() )
self.run = run
2016-11-23 14:05:38 +01:00
self.results = collections.OrderedDict()
2016-05-12 15:02:42 +02:00
self.swap = (swapx,swapy)
#self.clearCache()
2016-11-23 14:05:38 +01:00
d = self.data
2016-05-12 15:02:42 +02:00
self.spec1 = d.spec1 ; # spec1 is the one that is moved
self.spec2 = d.spec2 ;
self.E = alignment.defaultE
2016-05-12 15:02:42 +02:00
try:
self.loadTransform(initAlign)
except (AttributeError,FileNotFoundError):
if initAlign is None:
print("Set to default transform")
self.initAlign = self.setDefaultTransform()
else:
self.initAlign = initAlign
2016-05-12 15:02:42 +02:00
def getShots(self,shots=0,calib=None,bkgSub="line",roi=g_roi_height):
if shots == "all":
if calib != None:
shots = slice(0,self.nShotsPerCalib[calib])
else:
shots = slice(0,self.data.spec1.nShots)
2016-05-12 15:02:42 +02:00
# read data
im1 = self.spec1.getShots(shots,calib=calib)
im2 = self.spec2.getShots(shots,calib=calib)
2016-05-12 15:02:42 +02:00
# subtractBkg bkg
im1 = alignment.subtractBkg(im1,bkg_type=bkgSub)
im2 = alignment.subtractBkg(im2,bkg_type=bkgSub)
# rebin and swap im1 if necessary
if im1.shape[-1] != 1024:
im1 = mc.rebin(im1, (im1.shape[0],im1.shape[1],1024) )
2017-06-07 14:13:59 +02:00
if self.swap[0]: im1 = im1[:,:,::-1]
if self.swap[1]: im1 = im1[:,::-1,:]
2016-05-12 15:02:42 +02:00
if roi is None:
pass
elif isinstance(roi,slice):
im1 = im1[:,roi,:]
im2 = im2[:,roi,:]
elif isinstance(roi,int):
if not hasattr(self,"roi1"): self.roi1 = alignment.findRoi(im1[0],roi)
if not hasattr(self,"roi2"): self.roi2 = alignment.findRoi(im2[0],roi)
im1 = im1[:,self.roi1,:]; im2 = im2[:,self.roi2,:]
return im1,im2
def guiAlign(self,shot=0,save="auto"):
im1,im2 = self.getShot(shot)
gui = alignment.GuiAlignment(im1[0],im2[0])
input("Enter to start")
gui.start()
if save == "auto":
fname = g_folder_init+"/run%04d_gui_align.npy" % self.run
else:
fname = save
self.initAlign = gui.transform
gui.save(fname)
def doShot(self,shot=0,calib=None,initpars=None,im1=None,im2=None,doFit=True,show=False,showInit=False,save=False,savePlot="auto"):
if initpars is None: initpars= self.initAlign
if (im1 is None) or (im2 is None):
im1,im2 = self.getShots(shot,calib=calib); im1=im1[0]; im2=im2[0]
r = alignment.doShot(im1,im2,initpars,doFit=doFit,show=showInit)
im1 = r.im1
im2 = r.im2
self.initAlign = r.final_pars
if show:
if savePlot == "auto":
if not os.path.isdir(g_folder_out): os.makedirs(g_folder_out)
savePlot = g_folder_out+"/run%04d_calib%s_shot%04d_fit.png" % (self.run,calib,shot)
alignment.plotShot(im1,im2,res=r,save=savePlot)
if save: self.saveTransform()
return r
def doShots(self,shots=slice(0,50),calib=None,initpars=None,doFit=False,returnBestTransform=False,nSaveImg=5,nInChunks=250):
"""
shots : slice to define shots to read, use 'all' for all shots in calibcycle
nSaveImg : save saveImg images in memory (self.results), use 'all' for all
useful for decreasing memory footprint
"""
if initpars is None: initpars= self.initAlign
if shots == "all": shots = list(range(self.nShotsPerCalib[calib]))
if isinstance(shots,slice):
nmax = self.nShotsPerCalib[calib] if calib is not None else self.data.spec1.nShots
shots = sliceToIndices(shots,nmax)
chunks = x3py.toolsVarious.chunk(shots,nInChunks)
ret_chunks = []
for chunk in chunks:
s1,s2 = self.getShots(chunk,calib=calib)
ret_chunk = alignment.doShots(s1,s2,initpars=initpars,doFit=doFit,\
returnBestTransform=False,nSaveImg=nSaveImg)
ret_chunks.append(ret_chunk)
if nSaveImg != 'all' and len(chunk)<nSaveImg: nSaveImg = 0
ret = alignment.unravel_results(ret_chunks)
bestT,fom = alignment.getBestTransform(ret)
if doFit: self.initAlign = bestT
# keep it for later !
self.results[calib] = ret
if returnBestTransform:
return ret,bestT
else:
return ret
2016-05-12 15:02:42 +02:00
def analyzeScan(self,initpars=None,nShotsPerCalib="all",calibs="all",calibsToFit="all",nImagesToFit=0,nSaveImg=5):
""" nImagesToFit: number of images to Fit per calibcycle, (int or "all")
calibs to fit could be 'all','even','odd'
"""
2016-05-12 15:02:42 +02:00
if initpars is None: initpars= self.initAlign
if calibs == "all": calibs=list(range(self.nCalib))
if isinstance(calibs,slice): calibs=list(range(self.nCalib))[calibs]
nC = len(calibs)
for ic,calib in enumerate(calibs):
shots = list(range(self.nShotsPerCalib[calib]))
if nShotsPerCalib != "all": shots = shots[:nShotsPerCalib]
if nImagesToFit == "all":
nToFit = self.nShotsPerCalib[calib]
2016-11-23 17:16:39 +01:00
else:
nToFit = nImagesToFit
if calibsToFit == 'even' and (calib%2==1): nToFit=0
if calibsToFit == 'odd' and (calib%2==0): nToFit=0
print("Calib %d, tofit %d"%(calib,nToFit))
ret = None
if nToFit > 0:
ret,bestTransf = self.doShots(shots=shots[:nToFit],calib=calib,doFit=True,\
2016-11-23 17:16:39 +01:00
initpars=initpars,nSaveImg=nSaveImg,returnBestTransform=True);
initpars = bestTransf; self.initAlign=bestTransf
if nToFit < len(shots):
ret2 = self.doShots(shots[nToFit:],calib=calib,initpars=initpars,
doFit=False,nSaveImg=0)
if ret is None:
ret = ret2
else:
ret = alignment.unravel_results( (ret,ret2) )
#print("Memory available 3",x3py.toolsOS.memAvailable())
self.results[calib] = ret
#print("Memory available 4",x3py.toolsOS.memAvailable())
print("Calib cycle %d/%d -> %.3f (best FOM: %.2f)" % (ic,nC,self.scanpos[ic],np.nanmin(ret.fom)))
return [self.results[c] for c in calibs]
2016-05-12 15:02:42 +02:00
def save(self,fname="auto",overwrite=False):
if len(self.results) == 0:
print("self.results are empty, returning without saving")
2017-06-07 14:13:59 +02:00
return
2016-11-23 14:05:38 +01:00
if not os.path.isdir(g_folder_out): os.makedirs(g_folder_out)
2016-05-12 15:02:42 +02:00
if fname == "auto":
fname = g_folder_out+"/run%04d_analysis.npz" % self.run
2016-05-12 15:02:42 +02:00
if os.path.exists(fname) and not overwrite:
print("File %s exists, **NOT** saving, use overwrite=True is you want ..."%fname)
return
h = dict()
h["roi1"] = (self.roi1.start,self.roi1.stop)
h["roi2"] = (self.roi2.start,self.roi2.stop)
if hasattr(self.data.scan,"scanmotor0"):
h["scanmot0"] = self.data.scan.scanmotor0
else:
h["scanmot0"] = 'notascan'
h["scanpos0"] = self.data.scan.scanmotor0_values
if hasattr(self.data.scan,"scanmotor1"):
h["scanmot1"] = self.data.scan.scanmotor1
h["scanpos1"] = self.data.scan.scanmotor1_values
h["results"] = self.results
h["E"] = self.E
np.savez(fname,**h)
#h["transform"] = self.initAlign
2016-05-12 15:02:42 +02:00
def load(self,fname="auto"):
if fname == "auto": fname = g_folder_out+"/run%04d_analysis.npz" % self.run
temp = np.load(fname)
self.results = temp["results"].item()
temp.close()
2016-05-12 15:02:42 +02:00
def _auto_transform_name(self,run=None,calib=None):
if run is None: run = self.run
fname = g_folder_init+"/run%04d_transform" % run
if calib is not None:
fname = fname + "_c%03d" % calib
return fname + ".npy"
def saveTransform(self,fname="auto",calib=None,transform=None):
2016-05-12 15:02:42 +02:00
if transform is None: transform = self.initAlign
if fname == "auto": fname = self._auto_transform_name(calib=calib)
2016-05-12 15:02:42 +02:00
print("Saving roi and transformation parameter to %s"%fname)
alignment.saveAlignment(fname,self.initAlign,self.roi1,self.roi2,self.swap)
2016-05-12 15:02:42 +02:00
def loadTransform(self,fname="auto", calib=None):
if isinstance(fname,dict): raise FileNotFoundError
if fname == "auto": fname = self._auto_transform_name(calib=calib)
if isinstance(fname,int):
2016-05-12 15:02:42 +02:00
fname = g_folder_init+"/run%04d_transform.npy" % fname
if not os.path.exists(fname): print("Asked to read %s, but it does not exist"%fname)
2016-05-12 15:02:42 +02:00
temp = np.load(fname).item()
self.initAlign = temp["transform"]
self.roi1 = temp["roi1"]
self.roi2 = temp["roi2"]
if 'swap' in temp: self.swap=temp['swap']
2016-05-12 15:02:42 +02:00
print("init transform and ROIs from %s"%fname)
def clearCache(self):
del self.roi1
del self.roi2
alignment.clearCache(); # nedded for multiprocessing can leave bad parameters in the cache
def setDefaultTransform( self ):
#dict( scalex=0.65,rotation=0.0,transx=90, iblur1=4.3,fix_iblur1=False )
t = alignment.g_fit_default_kw
2016-05-12 15:02:42 +02:00
self.initAlign = t
return t
def quick_mec(run,ref=236,divideByRef=False,returnRes=False):
""" useful to analyze the runs around 140 (done with the focusing """
ref_run = 236
h=h5py.File("mecl3616_output/run%04d_analysis.h5" %ref,"r")
ref = np.nanmean(h["calibNone"]["ratio"][...],axis=0)
r = AnalyzeRun(run,initAlign=ref,swapx=True,swapy=False)
res=r.doShots(slice(5),doFit=False)
ret = res["ratio"]/ref if divideByRef else res["ratio"]
if returnRes:
return ret,res
else:
return ret
def quickAndDirty(run,nShots=300,returnAll=True,doFit=False):
""" useful to analyze the runs around 140 (done with the focusing """
r = AnalyzeRun(run,swap=True,initAlign=g_folder_init+"/run0144_transform.npy")
res=r.doShots(slice(nShots),doFit=doFit)
o = alignment.unravel_results(res)
ref = np.nanmedian(o["ratio"][:40],0)
sam = np.nanmedian(o["ratio"][50:],0)
if returnAll:
return sam/ref,o["ratio"]/ref
else:
return sam/ref