def ml_parameter_distributions(overwrite=False):
""" Compute distributions of microlensing event parameter fits by doing my MCMC
fit to all candidate, qso, not interesting, bad data, transient, and supernova
tagged light curves.
There are about ~4500
NEW: Hmm...maybe forget this, I keep breaking navtara
"""
ptf = mongo.PTFConnection()
light_curve_collection = ptf.light_curves
Nwalkers = 100
Nsamples = 1000
Nburn = 100
searched = []
max_parameters = []
for tag in ["candidate", "qso", "not interesting", "transient", "supernova", "bad data"]:
for lc_document in list(light_curve_collection.find({"tags" : tag})):
if str(lc_document["_id"]) in searched and not overwrite:
continue
light_curve = pdb.get_light_curve(lc_document["field_id"], lc_document["ccd_id"], lc_document["source_id"], clean=True)
sampler = fit.fit_model_to_light_curve(light_curve, nwalkers=Nwalkers, nsamples=Nsamples, nburn_in=Nburn)
max_idx = np.ravel(sampler.lnprobability).argmax()
# Turn this on to dump plots for each light curve
#fit.make_chain_distribution_figure(light_curve, sampler, filename="{0}_{1}_{2}_dists.png".format(lc_document["field_id"], lc_document["ccd_id"], lc_document["source_id"]))
#fit.make_light_curve_figure(light_curve, sampler, filename="{0}_{1}_{2}_lc.png".format(lc_document["field_id"], lc_document["ccd_id"], lc_document["source_id"]))
max_parameters.append(list(sampler.flatchain[max_idx]))
searched.append(str(lc_document["_id"]))
if len(searched) == 500:
break
if len(searched) == 500:
break
max_parameters = np.array(max_parameters)
fig, axes = plt.subplots(2, 2, figsize=(14,14))
for ii, ax in enumerate(np.ravel(axes)):
if ii == 3:
bins = np.logspace(min(max_parameters[:,ii]), max(max_parameters[:,ii]), 25)
ax.hist(max_parameters[:,ii], bins=bins, color="k", histtype="step")
else:
ax.hist(max_parameters[:,ii], color="k", histtype="step")
ax.set_yscale("log")
fig.savefig("plots/fit_events/all_parameters.png")
def fit_non_candidates(Nburn, Nsamples, Nwalkers):
# Now, the other ones...
non_candidates = [("PTFS1203am", "PTFS1212ak", "PTFS1213ap"),
("PTFS1215aj","PTFS1217ce", "PTFS1217df"),
("PTFS1217dk", "PTFS1219ad")]
for kk,row in enumerate(non_candidates):
#fig1 = plt.figure(figsize=(32,12))
fig1 = plt.figure(figsize=(22,8))
fontsize = 16
gs = gridspec.GridSpec(2, 3, height_ratios=[1,2])
# Make 6-panel figure with non-candidates
for ii,PTFID in enumerate(row):
if ii == 1 and kk == 2:
Nsamples = 10
#Nsamples = 1000
filename = "data/candidates/lc_{0}.dat".format(PTFID)
figtext = ptfid_to_longname[PTFID] #candidate["long_ptf_name"]
light_curve = lightcurve_from_filename(filename)
if ii == 0:
y_axis_labels = True
else:
y_axis_labels = False
if kk == 2 or (kk == 1 and ii == 2):
x_axis_labels = True
else:
x_axis_labels = False
logger.info("Fitting {0}".format(PTFID))
sampler = fit.fit_model_to_light_curve(light_curve, nwalkers=Nwalkers, nsamples=Nsamples, nburn_in=Nburn)
chain = sampler.flatchain
print "Mean acceptance fraction: {0:.3f}".format(np.mean(sampler.acceptance_fraction))
ax = plt.subplot(gs[1,ii])
ax2 = plt.subplot(gs[0,ii])
# "best" parameters
flatchain = np.vstack(sampler.chain)
best_m0, best_u0, best_t0, best_tE = np.median(flatchain, axis=0)
mjd = np.arange(light_curve.mjd.min()-1000., light_curve.mjd.max()+1000., 0.2)
mean_m0, mean_u0, mean_t0, mean_tE = np.median(chain,axis=0)
std_m0, std_u0, std_t0, std_tE = np.std(chain,axis=0)
print("m0 = ", mean_m0, std_m0)
print("u0 = ", mean_u0, std_u0)
print("tE = ", mean_tE, std_tE)
print("t0 = ", mean_t0, std_t0)
light_curve.mjd -= mean_t0
zoomed_light_curve = light_curve.slice_mjd(-3.*mean_tE, 3.*mean_tE)
if len(zoomed_light_curve) == 0:
mean_t0 = light_curve.mjd.min()
mean_tE = 10.
light_curve.mjd -= mean_t0
zoomed_light_curve = light_curve.slice_mjd(-3.*mean_tE, 3.*mean_tE)
zoomed_light_curve.plot(ax)
light_curve.plot(ax2)
ax.set_xlim(-3.*mean_tE, 3.*mean_tE)
#ax.text(-2.5*mean_tE, np.min(s_light_curve.mag), r"$u_0=${u0:.3f}$\pm${std:.3f}".format(u0=u0, std=std_u0) + "\n" + r"$t_E=${tE:.1f}$\pm${std:.1f} days".format(tE=mean_tE, std=std_tE), fontsize=19)
ylim, xlim = ax2.get_ylim(), ax2.get_xlim()
dx, dy = (xlim[1]-xlim[0]), (ylim[0]-ylim[1])
if (kk == 0 and ii == 1) or (kk == 1 and ii == 0) or (kk == 2 and ii == 0):
ax2.text((xlim[0]+xlim[1])/2.-15,ylim[1]+dy/8.,figtext,fontsize=fontsize)
else:
ax2.text(xlim[0]+dx/20., ylim[1]+dy/8., figtext,fontsize=fontsize)
if y_axis_labels:
ax.set_ylabel(r"$R$ (mag)", fontsize=fontsize+2)
if x_axis_labels:
ax.set_xlabel(r"time-$t_0$ [days]", fontsize=fontsize+2)
if y_axis_labels:
ax2.set_ylabel(r"$R$ (mag)", fontsize=fontsize+2)
fig1.subplots_adjust(hspace=0.15, wspace=0.15, left=0.06, right=0.95, top=0.98, bottom=0.09)
#fig1.savefig(os.path.join("plots/fit_events", "non_candidates_{0}.eps".format(kk)))
fig1.savefig(os.path.join("plots/fit_events", "fig14_{0}.eps".format(kk)))
"""
def fit_candidates(Nburn, Nsamples, Nwalkers, seed=None):
final_candidates = ["PTFS1206i", "PTFS1216bt", "PTFS1217cv"]
fig1 = plt.figure(figsize=(22,8))
fontsize = 16
gs = gridspec.GridSpec(2, 3, height_ratios=[1,2])
u0s,t0s,m0s,tEs = [],[],[],[]
du0s,dt0s,dm0s,dtEs = [],[],[],[]
# Make 3-panel figure with candidates
for ii,PTFID in enumerate(final_candidates):
if ii == 2:
Nsamples = 1000
chains = True
filename = "data/candidates/lc_{0}.dat".format(PTFID)
figtext = ptfid_to_longname[PTFID] #candidate["long_ptf_name"]
light_curve = lightcurve_from_filename(filename)
if ii == 0:
y_axis_labels = True
else:
y_axis_labels = False
if ii == 1:
x_axis_labels = True
else:
x_axis_labels = False
logger.info("Fitting {0}".format(PTFID))
sampler = fit.fit_model_to_light_curve(light_curve, nwalkers=Nwalkers, nsamples=Nsamples, nburn_in=Nburn, seed=seed)
chain = sampler.flatchain
print "Mean acceptance fraction: {0:.3f}".format(np.mean(sampler.acceptance_fraction))
ax = plt.subplot(gs[1,ii])
ax2 = plt.subplot(gs[0,ii])
# "best" parameters
flatchain = np.vstack(sampler.chain)
best_m0, best_u0, best_t0, best_tE = np.median(flatchain, axis=0)
mjd = np.arange(light_curve.mjd.min()-1000., light_curve.mjd.max()+1000., 0.2)
first_t0 = None
for jj in range(100):
walker_idx = np.random.randint(sampler.k)
chain = sampler.chain[walker_idx][-100:]
probs = sampler.lnprobability[walker_idx][-100:]
link_idx = np.random.randint(len(chain))
prob = probs[link_idx]
link = chain[link_idx]
m0, u0, t0, tE = link
if prob/probs.max() > 2:
continue
# More than 100% error
if np.any(np.fabs(link - np.mean(chain,axis=0)) / np.mean(chain,axis=0) > 0.25) or tE < 8 or tE > 250.:
continue
if np.fabs(t0 - best_t0) > 20.:
continue
if first_t0 == None:
first_t0 = t0
s_light_curve = SimulatedLightCurve(mjd=mjd, error=np.zeros_like(mjd), mag=m0)
s_light_curve.add_microlensing_event(t0=t0, u0=u0, tE=tE)
ax.plot(s_light_curve.mjd-first_t0, s_light_curve.mag, linestyle="-", color="#ababab", alpha=0.05, marker=None)
#ax_inset.plot(s_light_curve.mjd-first_t0, s_light_curve.mag, "r-", alpha=0.05)
mean_m0, mean_u0, mean_t0, mean_tE = np.median(chain,axis=0)
std_m0, std_u0, std_t0, std_tE = np.std(chain,axis=0)
m0s.append(mean_m0)
dm0s.append(std_m0)
t0s.append(mean_t0+54832.) # HJD to MJD
dt0s.append(std_t0)
tEs.append(mean_tE)
dtEs.append(std_tE)
u0s.append(mean_u0)
du0s.append(std_u0)
print("m0 = ", mean_m0, std_m0)
print("u0 = ", mean_u0, std_u0)
print("tE = ", mean_tE, std_tE)
print("t0 = ", mean_t0, std_t0)
light_curve.mjd -= mean_t0
zoomed_light_curve = light_curve.slice_mjd(-3.*mean_tE, 3.*mean_tE)
if len(zoomed_light_curve) == 0:
mean_t0 = light_curve.mjd.min()
mean_tE = 10.
light_curve.mjd -= mean_t0
zoomed_light_curve = light_curve.slice_mjd(-3.*mean_tE, 3.*mean_tE)
zoomed_light_curve.plot(ax)
light_curve.plot(ax2)
ax.set_xlim(-3.*mean_tE, 3.*mean_tE)
#ax.text(-2.5*mean_tE, np.min(s_light_curve.mag), r"$u_0=${u0:.3f}$\pm${std:.3f}".format(u0=u0, std=std_u0) + "\n" + r"$t_E=${tE:.1f}$\pm${std:.1f} days".format(tE=mean_tE, std=std_tE), fontsize=19)
ylim, xlim = ax.get_ylim(), ax.get_xlim()
dx, dy = (xlim[1]-xlim[0]), (ylim[0]-ylim[1])
ax.text(xlim[0]+dx/20., ylim[1]+dy/8.,
r"$u_0=${u0:.3f}$\pm${std:.3f}".format(u0=mean_u0, std=std_u0) +
"\n" + r"$t_E=${tE:.1f}$\pm${std:.1f} days".format(tE=mean_tE, std=std_tE), fontsize=fontsize)
ylim, xlim = ax2.get_ylim(), ax2.get_xlim()
dx, dy = (xlim[1]-xlim[0]), (ylim[0]-ylim[1])
if ii == 2:
ax2.text((xlim[0]+xlim[1])/2.-30,ylim[1]+dy/8.,figtext,fontsize=fontsize)
else:
ax2.text(xlim[0]+dx/20., ylim[1]+dy/8., figtext,fontsize=fontsize)
if y_axis_labels:
ax.set_ylabel(r"$R$ (mag)", fontsize=fontsize+2)
if x_axis_labels:
ax.set_xlabel(r"time-$t_0$ [days]", fontsize=fontsize+2)
# Now plot the "best fit" line in red
s_light_curve = SimulatedLightCurve(mjd=mjd, error=np.zeros_like(mjd), mag=best_m0)
s_light_curve.add_microlensing_event(t0=best_t0, u0=best_u0, tE=best_tE)
ax.plot(s_light_curve.mjd-best_t0, s_light_curve.mag, "k-", alpha=0.6, linewidth=3)
if y_axis_labels:
ax2.set_ylabel(r"$R$ (mag)", fontsize=fontsize+2)
import astropy.table as at
from astropy.io import ascii
tbl = at.Table([at.Column(data=u0s, name="u0"),
at.Column(data=du0s, name="e_u0"),
at.Column(data=tEs, name="tE"),
at.Column(data=dtEs, name="e_tE"),
at.Column(data=t0s, name="t0"),
at.Column(data=dt0s, name="e_t0"),
at.Column(data=m0s, name="m0"),
at.Column(data=dm0s, name="e_m0")])
ascii.write(tbl, output=os.path.join("plots/fit_events/metadata.tex"), Writer=ascii.Latex)
fig1.subplots_adjust(hspace=0.15, wspace=0.15, left=0.06, right=0.95, top=0.98, bottom=0.09)
fig1.savefig(os.path.join("plots/fit_events", "fig10.eps"))
