104 lines
3.7 KiB
Python
104 lines
3.7 KiB
Python
import sys
|
||
sys.path.insert(0,"../../../")
|
||
import collections
|
||
import os
|
||
import numpy as np
|
||
import matplotlib.pyplot as plt
|
||
|
||
import dispersiveXanes_utils as utils
|
||
import xppl37_spectra
|
||
import xanes_analyzeRun
|
||
import trx
|
||
import datastorage as ds
|
||
|
||
|
||
nice_colors = ["#1b9e77", "#d95f02", "#7570b3"]
|
||
nice_colors = "#1f78b4 #a6cee3 #b2df8a #33a02c".split()
|
||
gradual_colors = ['#014636', '#016c59', '#02818a', '#3690c0', '#67a9cf', '#a6bddb', '#d0d1e6']#, '#ece2f0']
|
||
|
||
def get_data(run,threshold=0.02,force=False):
|
||
if run == 80:
|
||
refCalibs=slice(1,None,2)
|
||
elif run == 84:
|
||
refCalibs=slice(None,None,2)
|
||
elif run == 76:
|
||
refCalibs=slice(None,None)
|
||
else:
|
||
refCalibs=slice(None,None,2)
|
||
fname = "../data/fig_fel_modes_run%04d.h5" % run
|
||
if not os.path.isfile(fname) or force:
|
||
r = xanes_analyzeRun.AnalyzeRun(run=run)
|
||
r.load()
|
||
E = r.E
|
||
calibs = list(r.results.keys())
|
||
calibs.sort()
|
||
calibs = calibs[refCalibs]
|
||
p1 = np.vstack( [r.results[c].p1 for c in calibs] )
|
||
p2 = np.vstack( [r.results[c].p2 for c in calibs] )
|
||
#temp = ds.DataStorage( E=E,p1=p1.astype(np.float16),p2=p2.astype(np.float16))
|
||
temp = ds.DataStorage( E=E,p1=p1,p2=p2)
|
||
_,_,Abs = xppl37_spectra.calcAbs( temp, threshold=0.02 )
|
||
temp.Abs = Abs #.astype(np.float16)
|
||
temp.info="Abs calculated with threshold = 0.02"
|
||
temp.save(fname)
|
||
data = ds.read(fname)
|
||
# nan is saved as -1 for masked arrays
|
||
for k in data.keys():
|
||
try:
|
||
data[k][data[k]==-1] =np.nan
|
||
except (TypeError,AttributeError):
|
||
pass
|
||
print("Run %d → nshots = %d"%(run,len(data.p1)))
|
||
p1,p2,Abs = xppl37_spectra.calcAbs( data, threshold=threshold )
|
||
data.Abs = Abs
|
||
return data
|
||
|
||
|
||
def fig_fel_modes(shots_per_run = slice(10,13),showAv=True,force=False,threshold=0.02,smootWidth=0.3):
|
||
runs = [28,39,54,76,80,84]
|
||
runs = [80,76,84]
|
||
#runs = [28,54,76,80,84]
|
||
figsize = [6,8]
|
||
fig,axes = plt.subplots( len(runs),2 , sharex=True, sharey=True,figsize=figsize)
|
||
#fig,axes = plt.subplots( 2,len(runs) , sharex=True, sharey='row')
|
||
#axes = axes.T
|
||
for run,ax in zip(runs,axes):
|
||
data = get_data(run,threshold=threshold,force=force)
|
||
E = data.E
|
||
s2 = data.p2[shots_per_run]
|
||
s1 = data.p1[shots_per_run]
|
||
norm = s2.max()*1.1
|
||
if showAv: ax[0].fill_between(E,0,data.p2.mean(0)/norm,color='#d95f0e',alpha=0.4)
|
||
for ispectrum,(spectrum1,spectrum2,a) in enumerate(zip(s1,s2,data.Abs)):
|
||
c = nice_colors[ispectrum]
|
||
ax[0].axhline(ispectrum,ls='--',lw=0.5,color=c)
|
||
# ax[1].axhline(ispectrum,ls='--',lw=0.5,color=c)
|
||
ax[0].plot(E,spectrum1/norm+ispectrum,lw=2,color=c)
|
||
ax[1].axhline(0.25+ispectrum,ls='--',lw=1,color=c)
|
||
# smooth does not work with nan's...
|
||
#if smootWidth > 0:
|
||
# a = xppl37_spectra.smoothSpectra(E,a,res=smootWidth)[0]
|
||
ax[1].plot(E,a+0.25+ispectrum,lw=2,color=c)
|
||
noise = np.nanstd(a)
|
||
ax[1].text(7135,0.4+ispectrum,"σ = %.2f"%noise)
|
||
ax[0].set_ylabel("Spectrum (a.u.) (run %d)"%run)
|
||
ax[1].set_ylabel("Absorption (run %d)"%run)
|
||
ax[0].grid(color="0.8",lw=0.5)
|
||
ax[1].grid(axis='x',color="0.8",lw=0.5)
|
||
tosave = np.vstack( (data.p2.mean(0)/norm,s2/norm) )
|
||
trx.utils.saveTxt("../data/fig_fel_modes_run%04d_spectra.txt"%run,E,tosave,info="E average_spectrum spectra")
|
||
trx.utils.saveTxt("../data/fig_fel_modes_run%04d_abs.txt"%run,E,data.Abs[shots_per_run],info="# threshold = %.2f\n# E Abs"%threshold)
|
||
|
||
axes[0,0].set_yticks(())
|
||
ax[0].set_xlim(7070,7180)
|
||
ax[0].set_ylim(-0.2,3.2)
|
||
axes[-1,0].set_xlabel("Energy (keV)")
|
||
axes[-1,1].set_xlabel("Energy (keV)")
|
||
#plt.subplots_adjust(left=0.07,right=0.95
|
||
plt.tight_layout()
|
||
plt.savefig("fig_fel_modes.png",transparent=True,dpi=300)
|
||
plt.savefig("fig_fel_modes.pdf",transparent=True)
|
||
|
||
if __name__ == "__main__": fig_fel_modes()
|
||
|