def pre(self, event):
global l
if cp.plot_vers == 2:
if haveImCube: self.ax3.remove()
#fx.visual()
if cp.plot_vers == 3:
self.ax4.remove()
self.cbar2.remove()
self.axslider2.remove()
#fx.visual()
self.ax2setup()
if cp.plot_vers != 1:
cp.plot_vers = 1
if 'ix' not in globals():
global ix, iy
ix = spectra.shape[1] // 2
iy = spectra.shape[0] // 2
spec = np.array(spectra[iy][ix])
self.ax1.scatter(ix,
iy,
s=200,
marker='x',
c='white',
linewidth=2.5)
self.ax2_title.set_text('Pixel (%ix , %iy): Spectra' % (ix, iy))
self.curveSpec.set_ydata(spec)
param = fx.paramSliders()
#s = M2(freqs, *param)
self.curveM2.set_ydata(M2(freqs, *param))
self.curveM1.set_ydata(M1(freqs, *param[:3]))
self.curveLorentz.set_ydata(m2(freqs, *param[3:6]))
#l, = ax2.loglog(freqs, s, lw=1.5, color='red')
plt.text(0.05, 11.5,
"*Using parameters found in: '%s'" % param_dir)
self.axreload = plt.axes([0.83, 0.18, 0.05, 0.05])
self.axsaveFig = plt.axes([0.83, 0.11, 0.05, 0.05])
self.axreset = plt.axes([0.83, 0.04, 0.05, 0.05])
self.breload = Button(self.axreload, 'Reload')
self.breload.on_clicked(fx.reload)
self.bsaveFig = Button(self.axsaveFig, 'Save')
self.bsaveFig.on_clicked(fx.saveFig)
self.breset = Button(self.axreset, 'Reset')
self.breset.on_clicked(fx.reset)
def update2(self):
params = np.zeros((len(fx.axsliders)))
for i in range(len(fx.axsliders)):
params[i] = fx.fnsliders[i].val
fx.fnsliders[i].valtext.set_text(text[i] % params[i])
cp.curveM2.set_ydata(M2(freqs, *params))
cp.curveM1.set_ydata(M1(freqs, *params[:3]))
cp.curveLorentz.set_ydata(m2(freqs, *params[3:6]))
def timeSeries(self, event):
global slid_mask
if cp.plot_vers == 1:
for slider in fx.axsliders:
slider.remove()
for axesObj in [self.axreload, self.axsaveFig, self.axreset]:
axesObj.remove()
if cp.plot_vers == 2: pass # maybe just: if cp.plot_vers != 2:
else:
if cp.plot_vers == 3:
self.ax4.remove()
self.cbar2.remove()
self.axslider2.remove()
#fx.visual()
cp.plot_vers = 2
if 'ix' not in globals():
global ix, iy
ix = spectra.shape[1]//2
iy = spectra.shape[0]//2
self.ax1.scatter(ix, iy, s=200, marker='x', c='white', linewidth=2.5)
self.ax2setup()
if haveImCube:
# ax3: displays pixel timeseries
self.ax3 = plt.subplot2grid((31,31),(22, 1), colspan=29, rowspan=8)
self.ax3_title = self.ax3.set_title('Pixel (%ix , %iy): Timeseries' % (ix, iy), fontsize=fontSize)
self.ax3.set_xlabel('Time', fontsize=fontSize, labelpad=5)
self.ax3.set_ylabel('Intensity', fontsize=fontSize, labelpad=5)
self.ax3.set_xlim(timestamps[0]-0.01*t_range, timestamps[-1]+0.01*t_range)
self.ts, = self.ax3.plot(timestamps, emptyTimeseries, 'k')
timeseries = np.array(imCube[iy+1][ix+1] / exposures)
self.ts.set_ydata(timeseries)
self.ax3.set_ylim(timeseries.min()*0.9, timeseries.max()*1.1)
spec = np.array(spectra[iy][ix])
self.ax2_title.set_text('Pixel (%ix , %iy): Spectra' % (ix, iy))
self.curveSpec.set_ydata(spec)
if haveParam:
cp.curveM2.set_ydata(M2(freqs, *h_map[iy,ix,:6]))
cp.curveM1.set_ydata(M1(freqs, *h_map[iy,ix,:3]))
cp.curveLorentz.set_ydata(m2(freqs, *h_map[iy,ix,3:6]))
def update2(val):
params = np.zeros((len(axsliders)))
for i in range(len(axsliders)):
params[i] = fnsliders[i].val
fnsliders[i].valtext.set_text(text[i] % params[i])
"""
if i == 0:
params[i] = 10**fnsliders[i].val
else:
params[i] = fnsliders[i].val
if i == 4:
fnsliders[i].valtext.set_text(text[i] % (1./(np.exp(params[4])*60.)))
else:
fnsliders[i].valtext.set_text(text[i] % params[i])
"""
s = M2(freqs, *params)
#l.set_ydata(s)
curveM2.set_ydata(s)
curveM1.set_ydata(M1(freqs, *params[:3]))
curveLorentz.set_ydata(m2(freqs, *params[3:6]))
def onclick(event):
#global ix, iy
ixx, iyy = event.xdata, event.ydata
if event.inaxes == cp.ax1:
global ix, iy
del cp.ax1.collections[:]
plt.draw()
if cp.spec_hist == 'hist':
cp.ax2setup()
cp.spec_hist = 'spec'
#print("location: (%fx, %fy)" % (ixx, iyy))
#print("location: (%ix, %iy)" % (ixx, iyy))
ix = int(round(ixx))
iy = int(round(iyy))
s = np.array(spectra[iy][ix])
if specVis_fit == True:
# use 3x3 pixel-box std.dev. or adhoc method for fitting uncertainties
if spec_unc == 'stddev':
ds = np.array(stdDev[iy][ix])
elif spec_unc == 'adhoc':
ds = ds0
specFit(s, ds)
# update subplots
cp.ax1.scatter(ix, iy, s=200, marker='x', c='white', linewidth=2.5)
cp.ax2_title.set_text('Pixel (%ix , %iy): Spectra' % (ix, iy))
cp.curveSpec.set_ydata(s)
# update spectra and timeseries
if cp.plot_vers == 2:
if haveImCube:
timeseries = np.array(imCube[iy+1][ix+1] / exposures)
cp.ts.set_ydata(timeseries)
cp.ax3.set_ylim(timeseries.min()*0.9, timeseries.max()*1.1)
cp.ax3_title.set_text('Pixel (%ix , %iy): Timeseries' % (ix, iy))
if haveParam:
cp.curveM2.set_ydata(M2(freqs, *h_map[iy,ix,:6]))
cp.curveM1.set_ydata(M1(freqs, *h_map[iy,ix,:3]))
cp.curveLorentz.set_ydata(m2(freqs, *h_map[iy,ix,3:6]))
# update spectra and model sliders
if cp.plot_vers == 1:
for slider in fx.axsliders:
slider.remove()
param = fx.paramSliders()
plt.text(0.05, 11.5, "*Using parameters found in: '%s'" % param_dir)
s = M2(freqs, *param)
#l.set_ydata(s)
cp.curveM2.set_ydata(s)
cp.curveM1.set_ydata(M1(freqs, *param[:3]))
cp.curveLorentz.set_ydata(m2(freqs, *param[3:6]))
def specFit(s, ds):
## fit data to combined power law plus gaussian component model
try:
m1_param = Fit(M1, freqs, s, p0=M1_guess, bounds=(M1_low, M1_high),
sigma=ds, method='dogbox')[0]
except RuntimeError: print("Error M1 - curve_fit failed")
except ValueError: print("Error M1 - inf/NaN")
A, n, C = m1_param # unpack model parameters
try:
m2_param0 = Fit(M2, freqs, s, p0=M2_guess, bounds=(M2_low, M2_high),
sigma=ds, method='dogbox', max_nfev=3000)[0]
except RuntimeError: print("Error M2 - curve_fit failed")
except ValueError: print("Error M2 - inf/NaN")
#A2, n2, C2, P2, fp2, fw2 = m2_param0
#print nlfit_gp
try:
m2_param = Fit(M2, freqs, s, p0=m2_param0, bounds=(M2_low, M2_high),
sigma=ds, max_nfev=3000)[0]
except RuntimeError: print("Error M2 - curve_fit failed")
except ValueError: print("Error M2 - inf/NaN")
A22, n22, C22, P22, fp22, fw22 = m2_param
#print m2_param
# create models from parameters
m1_fit = M1(freqs, *m1_param)
lorentz = m2(freqs, P22, fp22, fw22)
m2_fit2 = M2(freqs, *m2_param)
m1_fit2 = M1(freqs, A22, n22, C22)
"""
residsM1 = (s - m1_fit)
chisqrM1 = ((residsM1/ds)**2).sum()
residsM22 = (s - m2_fit2)
chisqrM22 = ((residsM22/ds)**2).sum()
redchisqrM22 = ((residsM22/ds)**2).sum()/float(freqs.size-6)
f_test2 = ((chisqrM1-chisqrM22)/(6-3))/((chisqrM22)/(freqs.size-6))
dof1, dof2 = 3, 6 # degrees of freedom for model M1, M2
#p_val = ff.sf(f_test2, dof1, dof2)
# extract the lorentzian amplitude scaling factor
amp_scale2 = M1(np.exp(fp22), A22, n22, C22)
"""
fwhm = (1. / (np.exp(fp22+fw22) - np.exp(fp22-fw22))) / 60.
pLoc = (1./np.exp(fp22))/60.
#curveM1A.set_ydata(m1_fit)
curveM2.set_ydata(m2_fit2)
curveM1.set_ydata(m1_fit2)
curveLorentz.set_ydata(lorentz)
p_index.set_text(r'$n$ = {0:0.2f}'.format(n22))
p_amp.set_text(r'$\alpha$ = {0:0.2e}'.format(P22))
p_loc.set_text(r'$\beta$ = {0:0.1f} [min]'.format(pLoc))
p_wid.set_text(r'$FWHM$ = {0:0.1f} [min]'.format(fwhm))
def onclick(event):
#global ix, iy
global fnsliders
global axsliders
ixx, iyy = event.xdata, event.ydata
if event.inaxes == ax1:
global ix, iy
del ax1.collections[:]
plt.draw()
#print("location: (%fx, %fy)" % (ixx, iyy))
#print("location: (%ix, %iy)" % (ixx, iyy))
ix = int(round(ixx))
iy = int(round(iyy))
#print(spectra)
s = np.array(spectra[iy][ix])
if specVis_fit == True:
# use 3x3 pixel-box std.dev. or adhoc method for fitting uncertainties
if spec_unc == 'stddev':
ds = np.array(stdDev[iy][ix])
elif spec_unc == 'adhoc':
ds = ds0
specFit(s, ds)
# update subplots
ax1.scatter(ix, iy, s=200, marker='x', c='white', linewidth=2.5)
title.set_text('Pixel (%ix , %iy): Spectra' % (ix, iy))
curveSpec.set_ydata(s)
timeseries = np.array(imCube[iy + 1][ix + 1] / exposures)
if toggle3 == 2:
ts.set_ydata(timeseries)
ax3.set_ylim(timeseries.min() * 0.9, timeseries.max() * 1.1)
title3.set_text('Pixel (%ix , %iy): Timeseries' % (ix, iy))
if haveParam:
curveM2.set_ydata(M2(freqs, *h_map[iy, ix, :6]))
curveM1.set_ydata(M1(freqs, *h_map[iy, ix, :3]))
curveLorentz.set_ydata(m2(freqs, *h_map[iy, ix, 3:6]))
if toggle3 == 1:
for slider in axsliders:
slider.remove()
axsliders = []
fnsliders = []
if haveParam:
param = h_map[iy, ix, :6]
else:
param = (np.array(M2_low) + np.array(M2_high)) / 2
# make parameter sliders
for i, M2_label in enumerate(M2_labels):
axsliders.append(plt.axes([0.15, 0.23 - (0.04 * i), 0.6,
0.02]))
fnsliders.append(
Slider(axsliders[i], M2_label, M2_low[i], M2_high[i],
param[i]))
fnsliders[i].on_changed(update2)
fnsliders[i].valtext.set_text(text[i] % param[i])
plt.text(0.05, 11.5,
"*Using parameters found in: '%s'" % param_dir)
s = M2(freqs, *param)
#l.set_ydata(s)
curveM2.set_ydata(s)
curveM1.set_ydata(M1(freqs, *param[:3]))
curveLorentz.set_ydata(m2(freqs, *param[3:6]))
def specFit(subcube, subcube_StdDev):
params = np.zeros((subcube.shape[0], subcube.shape[1], 9))
start_sub = timer()
T1 = 0
for l in range(subcube.shape[0]):
for m in range(subcube.shape[1]):
f = freqs
s = subcube[l][m]
# ---- use pixel-box std.dev. or adhoc method as fitting uncertainties
if spec_unc == 'stddev':
ds = subcube_StdDev[l][m]
elif spec_unc in ['adhoc', 'constant']:
ds = subcube_StdDev
# ---- fit models to spectra using SciPy's Levenberg-Marquart method
try:
m1_param = Fit(M1,
f,
s,
p0=M1_guess,
bounds=(M1_low, M1_high),
sigma=ds,
method='dogbox')[0]
except RuntimeError:
pass
except ValueError:
pass
# ---- first fit M2 model using 'dogbox' method
try:
m2_param0 = Fit(M2,
f,
s,
p0=M2_guess,
bounds=(M2_low, M2_high),
sigma=ds,
method='dogbox',
max_nfev=3000)[0]
except RuntimeError:
pass
except ValueError:
pass
# ---- next fit M2 model using default 'trf' method
try:
m2_param = Fit(M2,
f,
s,
p0=m2_param0,
bounds=(M2_low, M2_high),
sigma=ds,
max_nfev=3000)[0]
except RuntimeError:
pass
except ValueError:
pass
# ---- create model functions from fitted parameters
m1_fit = M1(f, *m1_param)
m2_fit = M2(f, *m2_param)
#weights = subcube_StdDev[l][m]
residsM1 = (s - m1_fit)
chisqrM1 = ((residsM1 / ds)**2).sum()
redchisqrM1 = chisqrM1 / float(f.size - 3)
residsM2 = (s - m2_fit)
chisqrM2 = ((residsM2 / ds)**2).sum()
redchisqrM2 = chisqrM2 / float(f.size - 6)
f_test = ((chisqrM1 - chisqrM2) / (6 - 3)) / ((chisqrM2) /
(f.size - 6))
# ---- extract the lorentzian-amplitude scaling factor
amp_scale = m2_param[3] / M1(np.exp(m2_param[4]), *m2_param[:3])
if chisqrM1 > chisqrM2:
# populate array with M2 parameters
params[l][m][:6] = m2_param
params[l][m][6] = f_test
params[l][m][7] = amp_scale
params[l][m][8] = redchisqrM2
else:
# populate array with M1 parameters
params[l][m][:3] = m1_param
params[l][m][3:8] = np.NaN
params[l][m][8] = redchisqrM1
# ---- estimate time remaining and print to screen
T = timer()
T2 = T - T1
if l == 0:
T_init = T - start_sub
T_est = T_init * (subcube.shape[0])
else:
T_est = T2 * (subcube.shape[0] - l)
T_min, T_sec = divmod(T_est, 60)
T_hr, T_min = divmod(T_min, 60)
if l == 0:
start_time = (T_hr, T_min, T_sec)
print("Thread %i on row %i/%i, ETR: %i:%.2i:%.2i" %
(rank, l, subcube.shape[0], T_hr, T_min, T_sec),
flush=True)
T1 = T
# ---- print estimated and total program time to screen
print("Beginning est. time = %i:%.2i:%.2i" % start_time, flush=True)
T_act = timer() - start_sub
T_min, T_sec = divmod(T_act, 60)
T_hr, T_min = divmod(T_min, 60)
print("Actual total time = %i:%.2i:%.2i" % (T_hr, T_min, T_sec),
flush=True)
return params