def plotPhi2(v_phi2, min_chi2, ndf, f_out, target):
import scipy.stats as st
from matplotlib import pyplot as plt
import matplotlib as mpl
plt.style.use('ggplot')
plt.figure(figsize=(8, 8))
goftarget = np.percentile(v_phi2, target)
goftarget68 = np.percentile(v_phi2, 68.8)
plt.hist(v_phi2,
np.linspace(min(v_phi2), max(v_phi2)),
alpha=0.6,
density=True,
label="Bootstrapped $\phi^2$")
_x = np.linspace(0, max(v_phi2), 1000)
plt.plot(_x, st.chi2.pdf(_x, ndf), label=r'$\chi^2(N_{df})=%i$' % (ndf))
plt.axvline(min_chi2, color="b", label="Tuning")
plt.axvline(goftarget,
color="k",
label="Target %.2f-th percentile of $\phi^2$" % target)
plt.axvline(goftarget68, color="m", label="68.8-th percentile of $\phi^2$")
plt.xlabel("$\phi^2$")
plt.axvline(ndf, label="$N_{df}=%i$" % (ndf))
plt.legend()
plt.savefigs(f_out)
plt.close()
def plotToyDef(xedges,
xmids,
yvals,
yerrs,
corr,
MIN,
MAX,
outfname="toy-defn"):
## Visualise the toy model
from matplotlib import pyplot as plt
import matplotlib as mpl
plt.style.use('ggplot')
plt.figure(figsize=(14, 5))
#
plt.subplot(121)
plt.fill_between(xmids, MIN, MAX, color="pink")
plt.errorbar(xmids,
yvals,
xerr=xedges[1:] - xmids,
yerr=yerrs,
marker="o",
linestyle="none",
color="k")
plt.yscale("log")
plt.xlabel("Observable")
plt.ylabel("Entries")
#
plt.subplot(122)
plt.imshow(corr,
interpolation="none",
norm=mpl.colors.Normalize(-1, 1),
cmap="RdBu_r")
#plt.matshow(cov)
plt.colorbar()
#
plt.savefigs(outfname)
#plt.show()
plt.close()
def plotChi2(v_chi2, v_chi2_full, ndf, f_out):
import scipy.stats as st
from matplotlib import pyplot as plt
import matplotlib as mpl
plt.style.use('ggplot')
plt.figure(figsize=(8, 8))
plt.hist(v_chi2,
np.linspace(min(v_chi2), max(v_chi2)),
alpha=0.6,
density=True,
label="Bootstrapped $\chi^2$ w/o correlation")
plt.hist(v_chi2_full,
np.linspace(min(v_chi2_full), max(v_chi2_full)),
alpha=0.6,
density=True,
label="Bootstrapped $\chi^2$ with correlation")
_x = np.linspace(0, max(v_chi2), 1000)
plt.plot(_x, st.chi2.pdf(_x, ndf), label=r'$\chi^2(N_{df})=%i$' % (ndf))
plt.axvline(ndf, label="$N_{df}=%i$" % (ndf))
plt.xlabel("$\chi^2$")
plt.legend()
plt.savefigs(f_out)
plt.close()
plt.hist(chi2s[1:],
bins=xchi2bins,
density=True,
histtype="stepfilled",
color="aliceblue",
edgecolor="darkblue",
hatch="",
label="Smear only",
alpha=0.7)
plt.axvline(chi2s_fit[0], ls="--", color="darkred", alpha=0.7)
plt.hist(chi2s_fit[1:],
bins=xchi2bins,
density=True,
histtype="stepfilled",
color="mistyrose",
edgecolor="darkred",
hatch=r"",
label="Smear+fit",
alpha=0.7)
plt.plot(xchi2s, [st.chi2.pdf(x, args.NDATABINS) for x in xchi2s],
color="darkblue",
label=r"$\chi^2(k = N_\mathrm{bin})$")
plt.plot(xchi2s, [st.chi2.pdf(x, args.NDATABINS - 2) for x in xchi2s],
color="darkred",
label=r"$\chi^2(k = N_\mathrm{bin}-2)$")
plt.xlabel(r"$\phi^2$")
plt.xlim(0, 2.5 * args.NDATABINS)
plt.legend()
plt.savefigs("simplechi2")
plt.show()
def plotEigenTuneHist2(jd, xmids, central, ETs, truth, plot_prefix=""):
TUNED = np.array(mkSignal2d(central, xmids))
ETD = [[
np.array(mkSignal2d(ETs[0], xmids)),
np.array(mkSignal2d(ETs[1], xmids))
],
[
np.array(mkSignal2d(ETs[2], xmids)),
np.array(mkSignal2d(ETs[3], xmids))
]]
# from IPython import embed
# embed()
dfp = []
for num, y in enumerate(TUNED):
a = max(map(abs, [ETD[0][0][num] - y, ETD[0][1][num] - y]))
b = max(map(abs, [ETD[1][0][num] - y, ETD[1][1][num] - y]))
dfp.append(np.sqrt(a**2 + b**2))
dfm = []
for num, y in enumerate(TUNED):
a = max(map(abs, [y - ETD[0][0][num], y - ETD[0][1][num]]))
b = max(map(abs, [y - ETD[1][0][num], y - ETD[1][1][num]]))
dfm.append(np.sqrt(a**2 + b**2))
DFP = TUNED + dfp
DFM = TUNED - dfm
ERR = np.sqrt(1. / np.array(jd["TUNEINVDATACOV"]).reshape(
(jd["NBINS"], jd["NBINS"])).diagonal())
YV = jd["YVALS"]
from matplotlib import pyplot as plt
import matplotlib as mpl
plt.style.use('ggplot')
plt.figure(figsize=(10, 8))
#
X = np.array(range(len(YV))) + 0.5
plt.fill_between(X, DFP, DFM, alpha=0.5, color="red")
plt.plot(X, truth, label="Truth", linewidth=3, color="black")
plt.yscale("log")
plt.xlabel("# Bin")
plt.ylabel("Entries")
etcols = ["b", "m"]
for num, ET in enumerate(ETD):
plt.plot(X,
ET[0],
"%s-" % etcols[num],
linewidth=3,
label="ET%i+" % (num + 1))
plt.plot(X,
ET[1],
"%s--" % etcols[num],
linewidth=3,
label="ET%i-" % (num + 1))
plt.errorbar(X,
YV,
yerr=ERR,
marker="o",
linestyle="none",
color="k",
label="'Data'")
plt.plot(X,
TUNED,
"r-",
linewidth=3,
label="Central tune ($\chi^2=%.1f$)" % jd["TUNECHI2"])
plt.legend()
plt.ylim((100, 580))
#
plt.savefigs("result-%s-%i-%s" %
(plot_prefix, jd["NBINS"], jd["CORRMODE"]))
plt.close()
iwin = np.argmin(allminPhi2)
winnerP = allminP[iwin]
winnerPhi2 = allminPhi2[iwin]
import matplotlib.pyplot as plt
plt.figure()
plt.hist(np.array(allallPhi2).flatten(),
100,
density=True,
color="red")
plt.axvline(winnerPhi2, c="black")
xchi2 = np.linspace(0, 50, 200)
fchi2 = st.chi2.pdf(xchi2, opts.NBINS - 1)
plt.plot(xchi2, fchi2, color="blue")
plt.savefigs(opts.PLOTPREFIX + "_phi2")
plt.close()
# This is the tuning main minimisation --- note that this uses the covariance WITHOUT sqrt(2)
Cinv_phi2 = np.diag([1. / e**2 for e in yerrs])
# Note add multistart here
MIN_tune = optimize.minimize(
lambda x: chi2wCov(yvals, xmids, Cinv_phi2, x), center, bounds=box)
Cinv_tune = MIN_tune["hess_inv"].todense()
jd = {
"CORRMODE": opts.CORRMODE,
"NBINS": opts.NBINS,
"NDF": opts.NBINS - 2, # TODO eventually nparams
"TUNECHI2": MIN_tune["fun"],