def test_truncated():
"""
this demonstrates that it doesn't work
"""
import ngmix
import biggles
nsamp=1000000
ptrunc=ngmix.priors.TruncatedSimpleGauss2D(0.0, 0.98, 0.4, 0.4, 1.0)
g1,g2 = ptrunc.sample(nsamp)
g=numpy.vstack( (g1,g2) ).T
gmm=mixture.GMM(n_components=1, covariance_type='full')
gmm.fit(g)
gs=gmm.sample(nsamp)
hc=biggles.make_histc(g[:,1], nbin=100, min=0.0, max=1.0, label='data', color='blue')
hcs=biggles.make_histc(gs[:,1], nbin=100, min=0.0, max=1.0, label='fit',color='red')
key=biggles.PlotKey(0.1,0.9,[hc,hcs], halign='left')
plt=biggles.FramedPlot(xlabel='g2')
plt.add( hc, hcs, key )
plt.show()
def _plot_single(data, samples, comps, do_ylog=False):
import biggles
import esutil as eu
import pcolors
valmin=data.min()
valmax=data.max()
std = data.std()
binsize=0.05*std
ph,be,harr = biggles.make_histc(data, min=valmin, max=valmax,
binsize=binsize,
ylog=do_ylog, norm=1,
get_hdata=True)
sample_ph,sbe,sharr= biggles.make_histc(samples, min=valmin, max=valmax,
binsize=binsize, color='red', ylog=do_ylog, norm=1,
get_hdata=True)
ph.label='data'
sample_ph.label='fit'
key = biggles.PlotKey(0.1, 0.9, [ph, sample_ph], halign='left')
plt = biggles.FramedPlot()
plt.add( ph, sample_ph, key )
w,=where( (harr > 0) & (sharr > 0) )
yrange=[min(harr[w].min(), sharr[w].min()),
1.1*max(harr[w].max(), sharr[w].max())]
if do_ylog:
plt.ylog=True
# now add the components
h,rev=eu.stat.histogram(comps, rev=True)
print(h)
w,=where(h > 0)
ncolors = w.size
colors=pcolors.rainbow(ncolors)
icolor=0
for i in xrange(h.size):
if rev[i] != rev[i+1]:
w=rev[ rev[i]:rev[i+1] ]
frac=float(w.size)/comps.size
ph = biggles.make_histc(samples[w],
#min=valmin, max=valmax,
binsize=binsize,
color=colors[icolor],
ylog=do_ylog,
norm=frac)
plt.add(ph)
icolor += 1
plt.yrange=yrange
return plt
def _plot_single(data, samples, do_ylog=False):
import biggles
valmin=data.min()
valmax=data.max()
std = data.std()
binsize=0.05*std
ph = biggles.make_histc(data, min=valmin, max=valmax, binsize=binsize, ylog=do_ylog, norm=1)
sample_ph= biggles.make_histc(samples, min=valmin, max=valmax, binsize=binsize, color='red', ylog=do_ylog, norm=1)
ph.label='data'
sample_ph.label='fit'
'''
hdict = get_norm_hist(data, min=valmin, max=valmax, binsize=binsize)
sample_hdict = get_norm_hist(samples, min=valmin, max=valmax, binsize=binsize)
hist=hdict['hist_norm']
sample_hist=sample_hdict['hist_norm']
if do_ylog:
hist=hist.astype('f8').clip(min=0.001)
sample_hist=sample_hist.astype('f8').clip(min=0.001)
ph = biggles.Histogram(hist, x0=valmin, width=4, binsize=binsize)
ph.label = 'data'
sample_ph = biggles.Histogram(sample_hist, x0=valmin,
width=1, color='red', binsize=binsize)
sample_ph.label = 'joint fit'
'''
key = biggles.PlotKey(0.1, 0.9, [ph, sample_ph], halign='left')
plt = biggles.FramedPlot()
plt.add( ph, sample_ph, key )
if do_ylog:
plt.ylog=True
#if do_ylog:
# plt.yrange=[0.1, 1.1*sample_hist.max()]
return plt
def compare_hist(data1, data2, names=None, dataset_names=None, nsig=10.0, **kw):
"""
Compare the normalized histograms for the two data sets. Make a grid of
plots if the data are multi-dimensional
parameters
----------
data1: array
a [N] or [N,dim] array
data2: array
a [M] or [M,dim] array
names: list, optional
Optional list of names for each dimension
dataset_names: list, optional
Optional list of names for each dataset
nsig: float, optional
Optional number of standard deviations to clip histograms,
default 10.0
"""
import biggles
from numpy import newaxis
from .stat import sigma_clip
if len(data1.shape)==1:
data1=data1[:,newaxis]
if len(data2.shape)==1:
data2=data2[:,newaxis]
n1,d1 = data1.shape
n2,d2 = data2.shape
if d1 != d2:
raise ValueError("data must have same number of dims. "
"got %d and %d" % (d1,d2))
if names is not None:
if len(names) != d1:
raise ValueError("names must have len equal to number of dims. "
" in data, got %d and %d" % (d1,len(names)))
else:
names = ['par%d' % i for i in xrange(d1)]
if dataset_names is not None:
if len(dataset_names) != 2:
raise ValueError("dataset_names must be len 2, "
"got %d" % len(dataset_names))
else:
dataset_names=['dataset1','dataset2']
if nsig is None:
nsig=100.0
grid = Grid(d1)
tab = biggles.Table(grid.nrow, grid.ncol)
pkw = {}
pkw.update(kw)
for dkeys in ['width','height']:
del pkw[dkeys]
pkw['visible']=False
for i in xrange(d1):
mn1,st1,ind1=sigma_clip(data1[:,i], nsig=nsig, get_indices=True)
mn2,st2,ind2=sigma_clip(data2[:,i], nsig=nsig, get_indices=True)
use_std = max(st1,st2)
binsize = 0.2*use_std
plt=biggles.FramedPlot()
plt.xlabel=names[i]
pkw['binsize']=binsize
pkw['color']='blue'
h1=biggles.make_histc(data1[ind1,i], **pkw)
pkw['color']='red'
h2=biggles.make_histc(data2[ind2,i], **pkw)
plt.add(h1,h2)
if i==0:
h1.label=dataset_names[0]
h2.label=dataset_names[1]
key=biggles.PlotKey(0.9,0.9,[h1,h2],halign='right')
plt.add(key)
row,col=grid(i)
tab[row,col] = plt
if 'show' in kw:
show=kw['show']
elif 'visible' in kw:
show=kw['visible']
else:
show=True
tab.aspect_ratio=kw.get('aspect_ratio',float(grid.nrow)/grid.ncol)
if show:
width=kw.get('width',1000)
height=kw.get('height',1000)
tab.show(width=width, height=height)
return tab
def plot_results(trials, **keys):
"""
Plot the points and histograms of trials from an MCMC chain.
"""
import biggles
import esutil
from esutil.numpy_util import where1, between
npars = trials.shape[1]
fontsize_min = keys.get("fontsize_min", 1)
biggles.configure("default", "fontsize_min", fontsize_min)
weights = keys.get("weights", None)
nbin = keys.get("nbin", 35)
names = keys.get("names", None)
show = keys.get("show", True)
nsigma = keys.get("nsigma", None)
means, cov = extract_stats(trials, weights=weights)
errs = sqrt(diag(cov))
plt = biggles.Table(npars, 2)
ind = numpy.arange(trials.shape[0])
for i in xrange(npars):
if names is not None:
name = names[i]
else:
name = r"$p_{%d}$" % i
use_trials = trials[:, i]
use_ind = ind
use_wts = weights
if nsigma is not None:
w = where1(between(trials[:, i], means[i] - nsigma * errs[i], means[i] + nsigma * errs[i]))
use_trials = trials[w, i]
use_ind = ind[w]
if weights is not None:
use_wts = weights[w]
burn_plot_i = biggles.plot(use_ind, use_trials, type="solid", xlabel="step", ylabel=name, visible=False)
plt[i, 0] = burn_plot_i
hcurve, bin_edges, harray = biggles.make_histc(use_trials, nbin=nbin, weights=use_wts, get_hdata=True)
plti = biggles.FramedPlot()
plti.xlabel = name
hmax = harray.max()
plti.yrange = [-0.05 * hmax, 1.2 * hmax]
plti.add(hcurve)
lab = r"$<%s> = %0.4g \pm %0.4g$" % (name, means[i], errs[i])
plab = biggles.PlotLabel(0.1, 0.8, lab, halign="left", color="blue")
plti.add(plab)
plt[i, 1] = plti
plt.title = keys.get("title", None)
if show:
plt.show()
return plt
def compare_hist(data1,
data2,
names=None,
dataset_names=None,
nsig=10.0,
**kw):
"""
Compare the normalized histograms for the two data sets. Make a grid of
plots if the data are multi-dimensional
parameters
----------
data1: array
a [N] or [N,dim] array
data2: array
a [M] or [M,dim] array
names: list, optional
Optional list of names for each dimension
dataset_names: list, optional
Optional list of names for each dataset
nsig: float, optional
Optional number of standard deviations to clip histograms,
default 10.0
"""
import biggles
from numpy import newaxis
from .stat import sigma_clip
if len(data1.shape) == 1:
data1 = data1[:, newaxis]
if len(data2.shape) == 1:
data2 = data2[:, newaxis]
n1, d1 = data1.shape
n2, d2 = data2.shape
if d1 != d2:
raise ValueError("data must have same number of dims. "
"got %d and %d" % (d1, d2))
if names is not None:
if len(names) != d1:
raise ValueError("names must have len equal to number of dims. "
" in data, got %d and %d" % (d1, len(names)))
else:
names = ['par%d' % i for i in xrange(d1)]
if dataset_names is not None:
if len(dataset_names) != 2:
raise ValueError("dataset_names must be len 2, "
"got %d" % len(dataset_names))
else:
dataset_names = ['dataset1', 'dataset2']
if nsig is None:
nsig = 100.0
grid = Grid(d1)
tab = biggles.Table(grid.nrow, grid.ncol)
pkw = {}
pkw.update(kw)
for dkeys in ['width', 'height']:
del pkw[dkeys]
pkw['visible'] = False
for i in xrange(d1):
mn1, st1, ind1 = sigma_clip(data1[:, i], nsig=nsig, get_indices=True)
mn2, st2, ind2 = sigma_clip(data2[:, i], nsig=nsig, get_indices=True)
use_std = max(st1, st2)
binsize = 0.2 * use_std
plt = biggles.FramedPlot()
plt.xlabel = names[i]
pkw['binsize'] = binsize
pkw['color'] = 'blue'
h1 = biggles.make_histc(data1[ind1, i], **pkw)
pkw['color'] = 'red'
h2 = biggles.make_histc(data2[ind2, i], **pkw)
plt.add(h1, h2)
if i == 0:
h1.label = dataset_names[0]
h2.label = dataset_names[1]
key = biggles.PlotKey(0.9, 0.9, [h1, h2], halign='right')
plt.add(key)
row, col = grid(i)
tab[row, col] = plt
if 'show' in kw:
show = kw['show']
elif 'visible' in kw:
show = kw['visible']
else:
show = True
tab.aspect_ratio = kw.get('aspect_ratio', float(grid.nrow) / grid.ncol)
if show:
width = kw.get('width', 1000)
height = kw.get('height', 1000)
tab.show(width=width, height=height)
return tab
def compare_hist(
data1,
data2,
names=None,
dataset_names=None,
nsig=10.0,
**kw,
):
"""
Compare the normalized histograms for the two data sets. Make a grid of
plots if the data are multi-dimensional
parameters
----------
data1: array
a [N] or [N,dim] array
data2: array
a [M] or [M,dim] array
names: list, optional
Optional list of names for each dimension
dataset_names: list, optional
Optional list of names for each dataset
nsig: float, optional
Optional number of standard deviations to clip histograms,
default 10.0
"""
import biggles
from numpy import newaxis
from .stat import sigma_clip
if len(data1.shape) == 1:
data1 = data1[:, newaxis]
if len(data2.shape) == 1:
data2 = data2[:, newaxis]
n1, d1 = data1.shape
n2, d2 = data2.shape
if d1 != d2:
raise ValueError("data must have same number of dims. "
"got %d and %d" % (d1, d2))
if names is not None:
if len(names) != d1:
raise ValueError("names must have len equal to number of dims. "
" in data, got %d and %d" % (d1, len(names)))
else:
names = ["par%d" % i for i in range(d1)]
if dataset_names is not None:
if len(dataset_names) != 2:
raise ValueError("dataset_names must be len 2, "
"got %d" % len(dataset_names))
else:
dataset_names = ["dataset1", "dataset2"]
if nsig is None:
nsig = 100.0
grid = Grid(d1)
tab = biggles.Table(grid.nrow, grid.ncol)
pkw = {}
pkw.update(kw)
for dkey in ["width", "height"]:
pkw.pop(dkey, None)
pkw["visible"] = False
pkw["norm"] = 1
if "nbin" not in pkw and 'binsize' not in pkw:
get_binsize = True
else:
get_binsize = False
for i in range(d1):
mn1, st1, ind1 = sigma_clip(data1[:, i], nsig=nsig, get_indices=True)
mn2, st2, ind2 = sigma_clip(data2[:, i], nsig=nsig, get_indices=True)
if get_binsize:
use_std = max(st1, st2)
pkw["binsize"] = 0.2 * use_std
plt = biggles.FramedPlot()
plt.xlabel = names[i]
pkw["color"] = "blue"
h1 = biggles.make_histc(data1[ind1, i], **pkw)
pkw["color"] = "red"
h2 = biggles.make_histc(data2[ind2, i], **pkw)
plt.add(h1, h2)
if i == 0:
h1.label = dataset_names[0]
h2.label = dataset_names[1]
key = biggles.PlotKey(0.9, 0.9, [h1, h2], halign="right")
plt.add(key)
row, col = grid(i)
tab[row, col] = plt
if "show" in kw:
show = kw["show"]
elif "visible" in kw:
show = kw["visible"]
else:
show = True
tab.aspect_ratio = kw.get("aspect_ratio", float(grid.nrow) / grid.ncol)
if show:
width = kw.get("width", 1000)
height = kw.get("height", 1000)
tab.show(width=width, height=height)
return tab
def multihist(data, binfac=0.1, **kw):
"""
plot a histogram for each dimension of the data
parameters
----------
data: array
array with shape [npoints, ndim]
binfac: float
The binsize for each dimension will be chosen as binfac*std(dimdata)
labels: optional
A sequence of labels for each dimension
"""
import biggles
if len(data.shape) != 2:
raise ValueError("data should have shape [npoints,ndim]")
ndim=data.shape[1]
labels=kw.pop('labels',None)
if labels is not None:
nl=len(labels)
assert len(labels)==ndim,"len(labels) = %d != %d" % (nl,ndim)
grid=Grid(ndim)
tab = kw.pop('plt',None)
if tab is not None:
add_to_existing_plots=True
else:
add_to_existing_plots=False
tab=biggles.Table(grid.nrow, grid.ncol)
tab.aspect_ratio=kw.pop('aspect_ratio',None)
if tab.cols != grid.ncol or tab.rows != grid.nrow:
m="input table has wrong dims. Expected %s got %s"
tup = ((grid.nrow,grid.ncol),(tab.rows,tab.cols))
raise ValueError(m % tup)
for dim in xrange(ndim):
ddata = data[:,dim]
mn, std = eu.stat.sigma_clip(ddata)
binsize=binfac*std
hc = biggles.make_histc(
ddata,
binsize=binsize,
**kw
)
row,col=grid(dim)
if add_to_existing_plots:
plt = tab[row,col]
plt.add(hc)
else:
plt = biggles.FramedPlot(aspect_ratio=1, **kw)
plt.add(hc)
tab[row,col]=plt
if labels is not None:
lab=labels[dim]
else:
lab='dim %d' % dim
tab[row,col].xlabel=lab
return tab
