def run_mcmc(self):
"""
Runs the MCMC fit
"""
if self.sampler is not None:
instance = None
prev_samples = self.sampler.iterations
else:
instance = self._setup_instance()
if self.fit_type == "both":
optres = self.run_downhill(instance)
if optres.success:
self.guess = list(optres.x)
prev_samples = 0
self._write_log_pre()
fitter = fit.MCMC(priors=self.priors,
data=self.y,
quantity=self.quantity,
constraints=self.constraints,
sigma=self.sigma,
guess=self.guess,
blobs=self.blobs,
verbose=self.verbose,
relax=self.relax)
start = time.time()
if self.chunks == 0:
self.chunks = self.nsamples - prev_samples
nchunks = np.ceil((self.nsamples - prev_samples) / float(self.chunks))
for i, s in enumerate(
fitter.fit(self.sampler, instance, self.nwalkers,
self.nsamples - prev_samples, self.burnin,
self.nthreads, self.chunks)):
# Write out files
self.write_iter_pickle(s)
print "Done {0}%. Time per sample: {1}".format(
100 * float(i + 1) / nchunks, (time.time() - start) /
((i + 1) * self.chunks * self.nwalkers))
total_time = time.time() - start
self._write_log_post(s, total_time)
self._write_data(s)
def sgr_rv_fits():
data = np.loadtxt("/home/newbym2/Dropbox/Research/sgrLetter/sgr_spec.csv", delimiter=",")
#dered_g,dered_r,l,b,ELODIERVFINAL,ELODIERVFINALERR
g0, r0 = data[:,0], data[:,1]
l, b = data[:,2], data[:,3]
rv, rv_err = data[:,4], data[:,5]
d = ac.getr(g0)
# Transform to vgsr from Yanny+ 2009
vgsr = ac.rv_to_vgsr(rv,l,b)
X,Y,Z, lsgr, bsgr, r_sgr = ac.lb2sgr(l, b, d)
for w in [0.5, 1.0, 2.5, 5.0, 7.5, 10.0, 15.0, 20.0, 25.0, 30.0, 50.0]:
keep = []
for i in range(len(data[:,0])):
if abs(bsgr[i]) < w:
if g0[i] < 20.0: continue
if g0[i] > 23.0: continue
keep.append(vgsr[i])
hist, edges = np.histogram(np.array(keep), bins=60, range=(-300.0, 300.0))
y, x = hist, edges[:-1]
e = func.poisson_errors(y)
fitter = fit.ToFit(x,y,e)
fitter.function=func.double_gaussian_one_fixed
fitter.update_params([3.0, -120.0, 30.0, 2.0, 0.0, 120.0])
fitter.step = [1.0, 10.0, 1.0, 1.0, 0.0, 0.0]
fitter.param_names = ["amp", "mu", "sigma", "amp", "mu", "sigma"]
path1 = fit.gradient_descent(fitter, its=10000, line_search=0)
path2 = fit.MCMC(fitter)
new_params=fitter.params
xx = sc.arange(-300.0, 300.0, 1.0)
yy = func.double_gaussian_one_fixed(xx, new_params)
y1 = func.gaussian_function(xx, new_params[:3])
y2 = func.gaussian_function(xx, new_params[3:])
fig = plt.figure()
plt.bar(edges[:-1], hist, width=10.0, color="white")
plt.plot(xx,yy, "k-")
plt.plot(xx,y1, "k--")
plt.plot(xx,y2, "k--")
plt.title("cut width = "+str(w))
plt.xlabel(r"$v_{\rm gsr}$", fontsize=16)
plt.ylabel(r"Counts")
#plt.ylim(0.0, 60.0)
#plt.show()
plt.savefig("/home/newbym2/Dropbox/Research/sgrLetter/sgr_spec/r_cut_relative"+str(w)+".png")
plt.close()
"""
fitter.function=func.double_gaussian
fitter.update_params([20.0, 0.0, 5.0, 20.0, 20.0, 15.0])
fitter.step = [1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
fitter.param_names = ["amp", "mu", "sigma", "amp", "mu", "sigma"]
fitter.function=func.quad_fat_gaussians
fitter.update_params([6.0, 10.0, 1.0, 6.0, -10.0, 1.0, 5.0, 10.0, 5.0, 10.0])
fitter.step = [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
fitter.param_names = ["amp", "mu", "sigma", "amp", "mu", "sigma", "amp","sigma", "amp","sigma"]
"""
print "\n {0}:".format(name)
print "# - Starting from:", fitter.params, fitter.RR
print "# - With steps:", fitter.step
path1 = fit.gradient_descent(fitter, its=10000, line_search=0)
path2 = fit.MCMC(fitter)
#errors = fit.get_Hessian_errors(fitter)
# cut from ToFit
new_params=fitter.params
plt.figure(1)
plt.bar(fitter.x, fitter.y, ((np.ma.max(fitter.x)-np.ma.min(fitter.x))/len(fitter.x)), align='center', color='w')
plt.plot(fitter.x, fitter.function(fitter.x, new_params), 'k-')
print new_params[:4]
plt.plot(fitter.x, func.gaussian_function(fitter.x, new_params[:3]), 'k--')
plt.plot(fitter.x, func.gaussian_function(fitter.x, new_params[3:6]), 'k--')
plt.plot(fitter.x, func.gaussian_function(fitter.x, [new_params[6], new_params[1], new_params[7]]), 'k--')
plt.plot(fitter.x, func.gaussian_function(fitter.x, [new_params[8], new_params[4], new_params[9]]), 'k--')
plt.xlabel("B")
plt.ylabel("counts")
plt.title(name, fontsize=8 )
out_name = name+"_plot"+".png"
print to_x params = pf.plane_OLS(to_x[:,0], to_x[:,1], to_x[:,2]) print "#- Distance array:", pf.plane_dist(to_x[:,0], to_x[:,1], to_x[:,2], params) fitter.x = to_x fitter.y = sc.zeros(len(data[:,0]) ) fitter.function=func.plane_fit #func.R_squared_plane fitter.update_params([3.0*ma.pi/2.0, ma.pi/2.0, 1.0]) #(params) fitter.range = [(0, 0.0, 2.0*ma.pi, ma.pi/10.0), (1, 0.0, ma.pi, ma.pi/20.0), (2, -20.0, 20.0, 1.0)] print fitter.params, fitter.RR fitter.step = [0.1, 0.1, 0.1] path1 = fit.MCMC(fitter, iters=10000, annealing=0, verbose=0) fitter.update_params([3.0*ma.pi/2.0, ma.pi/2.0, 1.0]) fitter.step = [0.75, 0.75, 0.75] print "# - Steps: ", fitter.step path2 = fit.gradient_descent(fitter, its=10000, line_search=0) errors = fit.get_Hessian_errors(fitter) #fitter.plot() yi, xi = 1,0 surface = fit.get_surface(fitter, yi, xi, plot=1, path=(path1[:,yi], path1[:,xi])) surface = fit.get_surface(fitter, yi, xi, plot=1, path=(path2[:,yi], path2[:,xi])) #print fitter.params, fitter.RR t, p, d = fitter.params dt, dp, dd = errors
def fit_hists(x_raw,
y_raw,
name,
outfile=None,
outpath="",
fit_type='double',
cuts=None,
background="common",
binsize=0.5,
lam_offset=0.0):
run = eval(name.split("_")[-1])
# clean data; don't fit bins with certain criteria
if cuts != None:
mask = sc.ones(len(y_raw), dtype=bool)
for i in range(cuts.shape[0]):
if cuts[i, 0] + lam_offset == run: #for offset Lambda wedges
if binsize == 0.5:
for j in range(cuts[i, 1], cuts[i, 2]):
mask[(j - 1)] = cuts[i, 3]
if binsize == 1.0:
for j in range(cuts[i, 1] / 2, cuts[i, 2] / 2):
mask[(j - 1)] = cuts[i, 3]
else:
mask = sc.zeros(len(y_raw), dtype=bool)
for i in range(len(y_raw)):
if y_raw[i] <= 0.0: mask[i] = 1 #flag zeros
#if y_raw[i] > 1000.0: mask[i] = 1 #flag globular clusters
# make new array that contains only mask==0
x, y = [], []
for i in range(len(y_raw)):
if mask[i] == 0:
x.append(x_raw[i])
y.append(y_raw[i])
# Set up data
x, y = np.array(x), np.array(y)
e = func.poisson_errors(y)
if background == "common":
# Make a flat background line set to most common bin height
hline, hedges = np.histogram(y, bins=40, range=(0.0, 1000.0))
best = 25.0 * np.argmax(hline) - 12.5
aa, bb = 0.0, best
elif background == "least":
ids = np.argsort(y)[:10]
aa, bb = 0.0, np.mean(y[ids])
elif background == "fitline":
# make a line, using the average of 10 bins on each end as the anchor points
x0, y0, x1, y1 = [], [], [], []
minbin = 1.0
if run < 120: i = 10
else: i = 1
while len(x0) < 10:
if y[i] > minbin:
x0.append(x[i])
y0.append(y[i])
i += 1
if run < 120: i = (len(y) / 2.0) - 5
elif run > 254: i = 65
else: i = -1
while len(x1) < 10:
if y[i] > minbin:
x1.append(x[i])
y1.append(y[i])
i -= 1
xi, yi = np.mean(x0), np.mean(y0)
xf, yf = np.mean(x1), np.mean(y1)
#xi, yi = np.mean(x[2:6]), np.mean(y[2:6])
#xf, yf = np.mean(x[-7:-3]), np.mean(y[-7:-3])
aa = (yf - yi) / (xf - xi)
bb = yf - (aa * xf)
else:
# null line
aa, bb = 0.0, 0.0
# fit it
fitter = fit.ToFit(x, y, e)
if fit_type == "double":
fitter.function = func.double_gauss_line
fitter.update_params([6.0, 0.0, 5.0, 6.0, -6.0, 5.0, aa, bb])
fitter.step = [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0]
fitter.param_names = [
"amp", "mu", "sigma", "amp", "mu", "sigma", "slope", "intercept"
]
elif fit_type == "quad":
fitter.function = func.quad_fat_gauss_line
fitter.update_params(
[6.0, 5.0, 1.0, 6.0, -10.0, 1.0, 5.0, 10.0, 5.0, 10.0, aa, bb])
fitter.step = [
1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 0.0
]
fitter.param_names = [
"amp", "mu", "sigma", "amp", "mu", "sigma", "amp", "sigma", "amp",
"sigma", "slope", "intercept"
]
elif fit_type == "triple":
if run < 120: tf, tm = 200.0, -50.0
if run > 120: tf, tm = 150.0, 40.0
fitter.function = func.triple_gauss_floor
fitter.update_params(
[6.0, 5.0, 1.0, 6.0, -10.0, 1.0, 2.0, tm, 40.0, tf])
fitter.step = [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 0.0, 5.0, 10.0]
fitter.param_names = [
"amp", "mu", "sigma", "amp", "mu", "sigma", "amp", "mu", "sigma",
"floor"
]
print "\n {0}:".format(name)
print "# - Starting from:", fitter.params, fitter.RR
print "# - With steps:", fitter.step
path1 = fit.gradient_descent(fitter, its=10000, line_search=0)
path2 = fit.MCMC(fitter)
#errors = fit.get_Hessian_errors(fitter)
# cut from ToFit
new_params = fitter.params
xx = np.arange(-70.0, 40.0, 0.1)
plt.figure(1)
plt.bar(x_raw, y_raw, binsize, align='center', color='white', zorder=1)
plt.bar(fitter.x,
fitter.y,
binsize,
align='center',
color='grey',
zorder=2)
plt.plot(xx, fitter.function(xx, new_params), 'k-')
print new_params[:4]
plt.plot(xx, func.gaussian_function(xx, new_params[:3]), 'k--')
plt.plot(xx, func.gaussian_function(xx, new_params[3:6]), 'k--')
if fit_type == "triple":
plt.plot(
xx,
func.gaussian_function(
xx, [new_params[6], new_params[7], new_params[8]]), 'k--')
plt.plot(xx, new_params[9] * sc.ones(len(xx)), "k--")
if fit_type == "quad":
plt.plot(
xx,
func.gaussian_function(
xx, [new_params[6], new_params[1], new_params[7]]), 'k--')
plt.plot(
xx,
func.gaussian_function(
xx, [new_params[8], new_params[4], new_params[9]]), 'k--')
plt.xlabel("B")
plt.ylabel("Counts")
plt.xlim(-70.0, 40.0)
plt.ylim(0.0, 700.0)
#plt.text(-65.0, 1100.0, name, fontsize=8 )
out_name = outpath + name + "_plot" + ".png"
#plt.show()
plt.savefig(out_name)
plt.close('all')
return np.insert(np.concatenate((fitter.params, fitter.error)), -1,
fitter.RR)