def sample_rp_ra_e_distributions(r,vr,chain,thin_by=100,burnin=200):
"""
sample rp,ra,epsilon distribution given chain and measurement
Arguments
---------
r: float
galactocentric radius of the dwarf
vr: float
the radial velocity of the dwarf
chain: array_like
mcmc chain of the potential parameters and k
thin_by: (=100) int
the thinning factor for the mcmc chain
burnin: (=200) int
the number of steps per walker to discard for burn in
Returns
-------
rp: array_like
posterior samples on rp
ra: array_like
posterior samples on ra
e: array_like
posterior samples on the eccentricity
"""
c = gu.reshape_chain(chain)[:,burnin::thin_by,:]
c = np.reshape(c, (c.shape[0]*c.shape[1],c.shape[2]))
k = c[:,2]
samples = c[:,-2:]
rp,ra,e = np.ones(len(samples))*np.nan,np.ones(len(samples))*np.nan,np.ones(len(samples))*np.nan
for i,params in enumerate(samples):
try:
vesc_Rsun,alpha = params
vesc = vesc_Rsun*(r/8.5)**(-alpha/2.)
v = vr + (vesc - vr)*(1. - np.random.power(k[i]+1) )
vT = np.sqrt(v**2. - vr**2.)
rp[i],ra[i] = get_rp_ra(r,vr,vT,params)
e[i] = (ra[i] - rp[i])/(ra[i] + rp[i])
except:
pass
return rp[~np.isnan(rp)],ra[~np.isnan(rp)],e[~np.isnan(rp)]
def Vesc_posterior(chain, model, burnin=200):
"""
Plot the posterior distribution on the escape speed as a function of radius.
Arguments
---------
chain: array_like[nsamples,ndims]
mcmc output
model: string
the name of the model to plot
burnin: int(=200)
number of steps from the default chain to disregard
"""
fig, ax = plt.subplots()
r = np.linspace(4., 50., 200)
n = m.get_numparams(model)
c = gu.reshape_chain(chain)[:, burnin:, :]
c = np.reshape(c, (c.shape[0] * c.shape[1], c.shape[2]))
samples = c[:, -n:].T
def vesc(r, params):
return m.vesc_model(r, 0., 0., params, model)
pl.posterior_1D(samples,
r,
vesc,
cmap="Blues",
ax=ax,
tickfontsize="small",
fontsize=mpl.rcParams['font.size'])
ax.set_xlabel("$r/\\mathrm{kpc}$")
ax.set_ylabel("$v_\\mathrm{esc}(r)/\\mathrm{kms^{-1}}$")
ax.errorbar(8.5,
533.,
yerr=[[41.], [54.]],
fmt='o',
c='k',
markersize=10.,
zorder=100000)
ax.text(8.9, 600., "P14", fontsize=25)
return fig, ax
def posterior_predictive_grid(v_grid,vmin,chain,model,tracer,burnin=200,pool_size=8):
"""
Compute the posterior predictive distribution given an MCMC chain and a model. Parallelise
over a given number of threads to speed up computation.
Arguments
---------
v_grid: array_like
an array of speeds at which to evaluate the posterior predictive distribution
vmin: float
the minimum speed considered
chain: array_like [nsamples,ndim]
an MCMC chain of model parameters
model: string
the name of the model
tracer: string
the type of tracer
burnin: int (=200)
the number of steps per walker to disregard as burn-in
pool_size: int (=8)
the size of the multiprocessing pool over which to distribute computation
Returns
-------
ppd: array_like
array of the same shape as v_grid, containing the posterior predictive probabilities
at each speed in v_grid
"""
#reshape the chain according to which model we're looking it
n = m.get_numparams(model)
c = gu.reshape_chain(chain)[:,burnin:,:]
c = np.reshape(c, (c.shape[0]*c.shape[1],c.shape[2]))
samples = c[:,-n:]
if tracer == "main_sequence":
k = c[:,2]
data = pd.read_csv("/data/aamw3/SDSS/main_sequence.csv")
lims,spline = construct_interpolator(data,"main_sequence")
elif tracer == "kgiant":
k = c[:,1]
data = pd.read_csv("/data/aamw3/SDSS/kgiant.csv")
lims,spline = construct_interpolator(data,"kgiant")
elif tracer == "bhb":
k = c[:,0]
data = pd.read_csv("/data/aamw3/SDSS/bhb.csv")
lims,spline = construct_interpolator(data,"bhb")
parfun = partial(posterior_predictive,vmin=vmin,k_samples=k,spline=spline,limits=lims\
,param_samples=samples,model=model)
pool = mp.Pool(pool_size)
output = pool.map(parfun,v_grid)
pool.close()
return output
def posterior_predictive_check(chain,tracer,model,vmin,nbins=20,thin_by=1,burnin=200):
"""
For every set of parameters in an MCMC chain, generate a mock data set of the same
size as the data.
Arguments
---------
chain: array_like [nsamples,ndim]
mcmc chain
tracer: string
type of tracer
model: string
the name of the model
vmin: float
the minimum speed considered
nbins: int (=20)
the number of bins in vLOS to use
thin_by: int(=1)
thin the chains by this factor
burnin: int(=200)
number of steps per walker to burn in
Returns
-------
bin_centres: array_like
centres of bins in vLOS
counts: array_like
the number counts of the data in each of the above bins
model_counts: array_like[nsamples,nstars]
the counts generated in each of the above bins in each mock sample
"""
n = m.get_numparams(model)
c = gu.reshape_chain(chain)[:,burnin::thin_by,:]
c = np.reshape(c, (c.shape[0]*c.shape[1],c.shape[2]))
samples = c[:,-n:]
if tracer == "main_sequence":
k = c[:,2]
data = pd.read_csv("/data/aamw3/SDSS/main_sequence.csv")
data = data[data.vgsr!=np.max(data.vgsr)].reset_index(drop=True) #remove the one outlier
elif tracer == "kgiant":
k = c[:,1]
data = pd.read_csv("/data/aamw3/SDSS/kgiant.csv")
else:
k = c[:,0]
data = pd.read_csv("/data/aamw3/SDSS/bhb.csv")
lims,spline = construct_interpolator(data,tracer)
data = data[np.abs(data.vgsr)>vmin].reset_index(drop=True)
N = len(data)
counts,bin_edges = np.histogram(np.abs(data.vgsr.values),nbins)
bin_centres = np.array([.5*(bin_edges[i] + bin_edges[i+1]) for i in np.arange(nbins)])
model_counts = np.zeros((len(k), nbins))
for i,theta in enumerate(samples):
v = draw_samples(N,vmin,k[i],spline,lims,theta,model)
model_counts[i,:],_ = np.histogram(v,bin_edges)
return bin_centres,counts,model_counts
def dwarf_galaxies(chain, model, burnin=200):
"""
Plot the dwarf galaxy line of sight velocities multiplied by sqrt(3) and our inference on the escape speed.
Arguments
---------
chain: array_like[nsamples,ndims]
mcmc output
model: string
the name of the model to plot
burnin: int(=200)
number of steps from the default chain to disregard
"""
fig, ax = plt.subplots()
dwarf_data = pd.read_csv("/data/aamw3/satellites/r_vgsr_dwarfs.csv")
r = np.linspace(np.min(dwarf_data['r']), np.max(dwarf_data['r']) + 10, 300)
n = m.get_numparams(model)
c = gu.reshape_chain(chain)[:, burnin:, :]
c = np.reshape(c, (c.shape[0] * c.shape[1], c.shape[2]))
samples = c[:, -n:].T
def vesc(r, params):
return m.vesc_model(r, 0., 0., params, model)
def m_vesc(r, params):
return -m.vesc_model(r, 0., 0., params, model)
pl.posterior_1D(samples,
r,
vesc,
cmap="Blues",
ax=ax,
tickfontsize="small",
fontsize=mpl.rcParams['font.size'])
pl.posterior_1D(samples,
r,
m_vesc,
cmap="Blues",
ax=ax,
tickfontsize="small",
fontsize=mpl.rcParams['font.size'])
ax.plot(dwarf_data['r'],
np.sqrt(3.) * dwarf_data['vgsr'],
'o',
mec='none',
ms=10,
c='0.5')
ax.plot(53, -np.sqrt(3.) * 211., 'o', ms=10, c='y', mec='none')
ax.plot(116., -np.sqrt(3.) * 189, 'o', ms=10, c='y', mec='none')
ax.plot(46., np.sqrt(3.) * 244, 'o', ms=10, c='r', mec='none')
ax.plot(37., -np.sqrt(3.) * 247, 'o', ms=10, c='r', mec='none')
ax.plot(126., np.sqrt(3.) * 154.3, 'o', ms=10, c='r', mec='none')
ax.annotate("Tuc 2",
xy=(53, -np.sqrt(3.) * 211.),
xytext=(75., -485.),
arrowprops=dict(facecolor='black', width=1,
shrink=0.15)) #tuc2
ax.annotate("Gru 1",
xy=(116., -np.sqrt(3.) * 189),
xytext=(130, -465.),
arrowprops=dict(facecolor='black', width=1,
shrink=0.15)) #gru1
ax.annotate("Boo III",
xy=(46., np.sqrt(3.) * 244),
xytext=(70, 480.),
arrowprops=dict(facecolor='black', width=1,
shrink=0.15)) #booIII
ax.annotate("Tri II",
xy=(37., -np.sqrt(3.) * 247),
xytext=(55., -564),
arrowprops=dict(facecolor='black', width=1,
shrink=0.15)) #triII
ax.annotate("Herc",
xy=(126., np.sqrt(3.) * 154.3),
xytext=(140., 440.),
arrowprops=dict(facecolor='black', width=1,
shrink=0.15)) #herc
ax.set_ylabel("$\sqrt{3}\,v_{||}/\\mathrm{kms^{-1}}$")
ax.set_xlabel("$r/\\mathrm{kpc}$")
return fig, ax
def halo_distribution(chain, model, burnin=200):
"""
Plot the mother sample line-of-sight velocity distributions and our inference on the escape speed.
Arguments
---------
chain: array_like[nsamples,ndims]
mcmc output
model: string
the name of the model to plot
burnin: int(=200)
number of steps from the default chain to disregard
"""
fig, ax = plt.subplots()
msto_data = pd.read_csv("/data/aamw3/SDSS/main_sequence.csv")
kgiant_data = pd.read_csv("/data/aamw3/SDSS/kgiant.csv")
bhb_data = pd.read_csv("/data/aamw3/SDSS/bhb.csv")
s_MSTO = gu.Ivesic_estimator(msto_data.g.values,msto_data.r.values\
,msto_data.i.values,msto_data.feh.values)
x_MSTO, y_MSTO, z_MSTO = gu.galactic2cartesian(s_MSTO, msto_data.b.values,
msto_data.l.values)
r_MSTO = np.sqrt(x_MSTO**2. + y_MSTO**2. + z_MSTO**2.)
s_BHB = gu.BHB_distance(bhb_data.g.values, bhb_data.r.values)
x_BHB, y_BHB, z_BHB = gu.galactic2cartesian(s_BHB, bhb_data.b.values,
bhb_data.l.values)
r_BHB = np.sqrt(x_BHB**2. + y_BHB**2. + z_BHB**2.)
r_halo = np.hstack((r_BHB, kgiant_data.rgc.values, r_MSTO))
v_halo = np.hstack(
(bhb_data.vgsr.values, kgiant_data.vgsr.values, msto_data.vgsr.values))
r = np.linspace(1., 50., 100)
n = m.get_numparams(model)
c = gu.reshape_chain(chain)[:, burnin:, :]
c = np.reshape(c, (c.shape[0] * c.shape[1], c.shape[2]))
samples = c[:, -n:].T
def vesc(r, params):
return m.vesc_model(r, 0., 0., params, model)
def m_vesc(r, params):
return -m.vesc_model(r, 0., 0., params, model)
pl.posterior_1D(samples,
r,
vesc,
cmap="Blues",
ax=ax,
tickfontsize="small",
fontsize=mpl.rcParams['font.size'])
pl.posterior_1D(samples,
r,
m_vesc,
cmap="Blues",
ax=ax,
tickfontsize="small",
fontsize=mpl.rcParams['font.size'])
ax.plot(r_halo, v_halo, 'o', ms=5, c='0.3', mec='none', rasterized=True)
ax.set_ylabel("$v_{||}/\\mathrm{kms^{-1}}$")
ax.set_xlabel("$r/\\mathrm{kpc}$")
ax.set_ylim((-600., 600.))
ax.axhline(-200., ls='--', c='k')
ax.axhline(+200., ls='--', c='k')
return fig, ax
def plot_mass_enclosed(chain,
model,
burnin=200,
cmap="Blues",
thin_by=1,
**kwargs):
"""
Plot the mass enclosed implied by a spherically symmetric model given
an MCMC chain of parameter samples and radial limits. Median and 68 percent,
94 percent credible intervals
Arguments
---------
chain: array_like [n_samples,n_parameters]
MCMC chain of parameter samples
model: string
model name
burnin: (=200) int
the number of steps per walker to discard as burn in
cmap: (="Blues") string
matpotlib colormap to use
thin_by: (=1) int
thinning factor for MCMC chains
Returns
-------
ax: pyplot.axes
matplotlib.pyplot axes object
"""
G = 43010.795338751527 #in km^2 s^-2 kpc (10^10 Msun)^-1
n = m.get_numparams(model)
c = gu.reshape_chain(chain)[:, burnin::thin_by, :]
c = np.reshape(c, (c.shape[0] * c.shape[1], c.shape[2]))
samples = c[:, -n:].T
rlims = [0.1, 70.]
if model == "spherical_powerlaw":
model_name = "SPL"
def mass_enclosed(r, params):
vesc_Rsun, alpha = params
v0 = np.sqrt(.5 * alpha) * vesc_Rsun
return v0**2. * r * (r / 8.5)**(-alpha) / G
elif model == "TF":
model_name = "TF"
def mass_enclosed(r, params):
v0, rs, alpha = params
return r * v0**2. * rs**alpha / (G *
(rs**2. + r**2.)**(.5 * alpha))
def rotation_curve(r, params):
return np.sqrt(mass_enclosed(r, params) * G / r)
r1 = np.linspace(rlims[0], rlims[1], 200)
r2 = np.linspace(8.5, rlims[1], 200)
width, height = plt.rcParams.get('figure.figsize')
fig, ax = plt.subplots(1, 2, figsize=(2 * width, height))
pl.posterior_1D(samples,
r1,
mass_enclosed,
cmap=cmap,
ax=ax[0],
tickfontsize="small",
fontsize=mpl.rcParams['font.size'],
**kwargs)
pl.posterior_1D(samples,
r2,
rotation_curve,
cmap=cmap,
ax=ax[1],
tickfontsize="small",
fontsize=mpl.rcParams['font.size'],
**kwargs)
ymin, ymax = ax[0].get_ylim()
ax[0].set_ylabel("$M(r)/\\mathrm{10^{10}M_\\odot}$")
ax[0].set_xlabel("$r/\\mathrm{kpc}$")
ax[1].set_xlabel("$r/\\mathrm{kpc}$")
ax[1].set_ylabel("$v_c(r)/\\mathrm{kms^{-1}}$")
ax[1].set_ylim((100., 350.))
colors = (Set1_6.mpl_colors[0], Set1_6.mpl_colors[2], Set1_6.mpl_colors[3],
Set1_6.mpl_colors[4], 'k')
#now plot other people's masses...
ax[0].errorbar(60.0,
40.0,
yerr=7.0,
fmt='o',
label="X08",
c=colors[0],
ms=9,
mec='none',
zorder=100,
elinewidth=2) #Xue 2008
ax[0].errorbar(50.0 + 1,
54.0,
yerr=[[36.0], [2.0]],
fmt='^',
label="W99",
c=colors[1],
ms=9,
mec='none',
zorder=100,
elinewidth=2) #Wilkinson 1999
ax[0].errorbar(50.0 - 1,
42.0,
yerr=4.0,
fmt='v',
label="D12",
c=colors[2],
ms=9,
mec='none',
zorder=100,
elinewidth=2) #Deason Broken Degneracies
ax[0].errorbar(50.0,
44.8,
yerr=[[1.4], [1.5]],
fmt='d',
label="WE15",
c=colors[3],
ms=9,
mec='none',
zorder=100,
elinewidth=2) #me and wyn 2015
ax[0].errorbar(50.0,
29.,
yerr=[[5.], [5.]],
fmt='s',
label="G14",
c=colors[4],
ms=9,
mec='none',
zorder=100,
elinewidth=2) #simon 2014
ax[0].legend(loc='best', numpoints=1)
#...and circular speeds
circspeed = lambda mass, radius: np.sqrt(mass * G / radius)
ax[1].errorbar( 60.0, circspeed(40.,60.), yerr = [[circspeed(40.,60.)-circspeed(40.-7.,60.)],[circspeed(40.+7.,60.)-circspeed(40.,60.)]],\
fmt='o', label = "X08", c=colors[0], ms=9, mec='none', zorder=100,elinewidth=2) #Xue 2008
ax[1].errorbar( 50.0 + 1, circspeed(54.,50.), yerr = [[circspeed(54.,50.)-circspeed(53.-36.,50.)],[circspeed(54+2.,50.)-circspeed(54.,50.)]],\
fmt='^', label = "W99", c=colors[1], ms=9, mec='none', zorder=100,elinewidth=2) #Wilkinson 1999
ax[1].errorbar( 50.0 - 1, circspeed(42.,50.), yerr = [[circspeed(42.,50.)-circspeed(42.-4.,50.)],[circspeed(42+4.,50.)-circspeed(42.,50.)]]\
, fmt='v', label = "D12", c=colors[2], ms=9, mec='none', zorder=100,elinewidth=2) #Deason Broken Degneracies
ax[1].errorbar(50.0, 198.2, yerr=[[3.2],[3.4]]\
, fmt='d', label = "WE15", c=colors[3], ms=9, mec='none', zorder=100,elinewidth=2) #me and wyn 2015
ax[1].errorbar(50.0, circspeed(29.,50.), yerr=[[circspeed(29.,50.)-circspeed(29.-5.,50.)],[circspeed(29+5.,50.)-circspeed(29.,50.)]]\
, fmt='s', label = "G14", c=colors[4], ms=9, mec='none', zorder=100,elinewidth=2) #simon 2014
return ax
def posterior_predictive_distribution(chain,
model,
burnin=200,
cmap="Greys",
thin_by=10,
nbins=[20, 20, 10],
pool_size=8):
"""
Plot the inferred line of sight velocity distribution for stars, normalised
to the escape velocity at their position.
Arguments
---------
chain: array_like
MCMC chain of parameters
model: string
name of model to be plotted
burnin: (=200) int
number of steps to discard as burn-in
Returns
-------
ax: matplotlib axes
axes object with plot
"""
major_formatter = pl.FuncFormatter(pl.my_formatter)
#reshape the chain according to which model we're looking it
n = m.get_numparams(model)
c = gu.reshape_chain(chain)[:, burnin::thin_by, :]
c = np.reshape(c, (c.shape[0] * c.shape[1], c.shape[2]))
samples = c[:, -n:]
#get the samples for k
k_bhb = c[:, 0]
k_kgiant = c[:, 1]
k_ms = c[:, 2]
k = [k_ms, k_kgiant, k_bhb]
tracer_names = ["main_sequence", "kgiant", "bhb"]
#load the data, extract vgsr and r from it
bhb = pd.read_csv("/data/aamw3/SDSS/bhb.csv")
kgiant = pd.read_csv("/data/aamw3/SDSS/kgiant.csv")
msto = pd.read_csv("/data/aamw3/SDSS/main_sequence.csv")
msto = msto[msto.vgsr != np.max(msto.vgsr)].reset_index(drop=True)
data = (msto, kgiant, bhb)
tracer_title = ["MSTO", "K-giant", "BHB"]
colors = ["k", "crimson", "royalblue"]
fig, ax = plt.subplots(3, 2, figsize=(10., 15.), sharex=True)
for a in ax.ravel():
a.yaxis.set_major_formatter(major_formatter)
cm = plt.cm.get_cmap(cmap)
v = np.linspace(200., 550., 100)
for i in np.arange(len(data)):
ppd = f.posterior_predictive_grid(v,
200.,
chain,
model,
tracer_names[i],
burnin=200,
pool_size=pool_size)
for a in ax[i, :]:
a.plot(v, ppd, c='slategray', zorder=0, lw=2)
a.set_xlim(np.min(v), np.max(v))
for i, tracer in enumerate(data):
tracer = tracer[np.abs(tracer.vgsr) > 200.].reset_index(drop=True)
counts, bins = np.histogram(np.abs(tracer.vgsr.values), nbins[i])
v_centres = np.array(
[.5 * (bins[j] + bins[j + 1]) for j in np.arange(nbins[i])])
nf = (len(tracer) * (bins[1] - bins[0]))**-1.
lowers, medians, uppers = np.zeros_like(v_centres), np.zeros_like(
v_centres), np.zeros_like(v_centres)
for j in np.arange(nbins[i]):
lowers[j] = gammaincinv(counts[j] + 1., 0.1) * nf
medians[j] = gammaincinv(counts[j] + 1., 0.5) * nf
uppers[j] = gammaincinv(counts[j] + 1., 0.9) * nf
ax[i, 0].errorbar(v_centres,
medians,
yerr=[medians - lowers, uppers - medians],
fmt='o',
ecolor=colors[i],
ms=4.,
mfc=colors[i],
mec='none',
elinewidth=1)
ax[i, 1].errorbar(v_centres,
medians,
yerr=[medians - lowers, uppers - medians],
fmt='o',
ecolor=colors[i],
ms=4.,
mfc=colors[i],
mec='none',
elinewidth=1)
ax[i, 1].set_yscale("log")
ymin, ymax = ax[i, 1].get_ylim()
ax[i, 1].set_ylim((1e-6, ymax))
ax[i, 0].text(400.,
0.75 * ax[i, 0].get_ylim()[1],
tracer_title[i],
fontsize=20)
fig.text(0.5, 0., "$v_{||}/\\mathrm{kms^{-1}}$")
fig.text(0.,
0.5,
"$p(v_{||}\\, | \\, \\mathrm{data})/\\mathrm{km^{-1}s}$",
rotation=90)
return fig, ax