def _summary(name, param, posterior, chi_sq):
"""
Find summary statistics for a single parameter.
:param name: Name of parameter
:type name: string
:param param: Data column of parameter
:type param:
:param posterior:
:type posterior:
:param chi_sq:
:type chi_sq:
:returns: List of summary statistics for a particular parameter
:rtype: list
"""
# Best-fit point
bestfit = stats.best_fit(chi_sq, param)
# Posterior mean
post_mean = stats.posterior_mean(posterior, param)
# Credible regions
pdf_data = one_dim.posterior_pdf(param,
posterior,
nbins=default("nbins"),
bin_limits=default("bin_limits")
)
lower_credible_region = one_dim.credible_region(pdf_data.pdf,
pdf_data.bin_centers,
alpha=default("alpha")[1],
region="lower")
upper_credible_region = one_dim.credible_region(pdf_data.pdf,
pdf_data.bin_centers,
alpha=default("alpha")[1],
region="upper")
summary = [name,
bestfit,
post_mean,
lower_credible_region,
upper_credible_region
]
return summary
def __init__(self, data, plot_options):
super(TwoDimPlot, self).__init__(data, plot_options)
opt = self.plot_options
# If the user didn't specify bin or plot limits,
# we find the extent of the data and use that to set them.
extent = np.zeros(4)
extent[0] = min(self.xdata)
extent[1] = max(self.xdata)
extent[2] = min(self.ydata)
extent[3] = max(self.ydata)
if self.plot_options.bin_limits is None:
self.plot_options = self.plot_options._replace(
bin_limits=[[extent[0], extent[1]], [extent[2], extent[3]]])
if self.plot_options.plot_limits is None:
self.plot_options = self.plot_options._replace(plot_limits=extent)
# Posterior PDF
if opt.kde_pdf:
# KDE estimate of PDF
self.pdf_data = two_dim.kde_posterior_pdf(
self.xdata,
self.ydata,
self.posterior,
bw_method=opt.bw_method,
bin_limits=opt.bin_limits)
else:
# Binned estimate of PDF
self.pdf_data = two_dim.posterior_pdf(self.xdata,
self.ydata,
self.posterior,
nbins=opt.nbins,
bin_limits=opt.bin_limits)
# Profile likelihood
self.prof_data = two_dim.profile_like(self.xdata,
self.ydata,
self.chisq,
nbins=opt.nbins,
bin_limits=opt.bin_limits)
# As with the 1D plots we use raw data for the best-fit point,
# and binned data for the mean and mode.
# Best-fit point
self.best_fit_x = stats.best_fit(self.chisq, self.xdata)
self.best_fit_y = stats.best_fit(self.chisq, self.ydata)
self.summary.append("Best-fit point (x,y): {}, {}".format(
self.best_fit_x, self.best_fit_y))
# Posterior mean
self.posterior_mean_x = stats.posterior_mean(
np.sum(self.pdf_data.pdf, axis=1), self.pdf_data.bin_centers_x)
self.posterior_mean_y = stats.posterior_mean(
np.sum(self.pdf_data.pdf, axis=0), self.pdf_data.bin_centers_y)
self.summary.append("Posterior mean (x,y): {}, {}".format(
self.posterior_mean_x, self.posterior_mean_y))
# Posterior mode
self.posterior_modes = two_dim.posterior_mode(*self.pdf_data)
self.summary.append("Posterior modes/s (x,y): {}".format(
self.posterior_modes))
# Posterior median
self.posterior_median_x = one_dim.posterior_median(
np.sum(self.pdf_data.pdf, axis=1), self.pdf_data.bin_centers_x)
self.posterior_median_y = one_dim.posterior_median(
np.sum(self.pdf_data.pdf, axis=0), self.pdf_data.bin_centers_y)
self.summary.append("Posterior median (x,y): {}, {}".format(
self.posterior_median_x, self.posterior_median_y))
def __init__(self, data, plot_options):
super(OneDimPlot, self).__init__(data, plot_options)
opt = self.plot_options
# If the user didn't specify bin or plot limits,
# we find the extent of the data and use that to set them.
extent = np.zeros(4)
extent[0] = min(self.xdata)
extent[1] = max(self.xdata)
extent[2] = 0
extent[3] = 1.2
# Downside of using named tuple is they're immutable
# so changing options is (justifiably) annoying.
# If this happens a lot (it shouldn't), consider
# using a mutable type instead...
if self.plot_options.bin_limits is None:
self.plot_options = self.plot_options._replace(
bin_limits=[extent[0], extent[1]])
if self.plot_options.plot_limits is None:
self.plot_options = self.plot_options._replace(plot_limits=extent)
# Posterior PDF. Norm by area if not showing profile likelihood,
# otherwise norm max value to one.
if opt.kde_pdf:
# KDE estimate of PDF
self.pdf_data = one_dim.kde_posterior_pdf(
self.xdata,
self.posterior,
bin_limits=opt.bin_limits,
norm_area=not opt.show_prof_like,
bw_method=opt.bw_method)
else:
# Binned estimate of PDF
self.pdf_data = one_dim.posterior_pdf(
self.xdata,
self.posterior,
nbins=opt.nbins,
bin_limits=opt.bin_limits,
norm_area=not opt.show_prof_like)
# Profile likelihood
self.prof_data = one_dim.prof_data(self.xdata,
self.chisq,
nbins=opt.nbins,
bin_limits=opt.bin_limits)
# Note the best-fit point is calculated using the raw data,
# while the mean, median and mode use the binned PDF.
# Best-fit point
self.best_fit = stats.best_fit(self.chisq, self.xdata)
self.summary.append("Best-fit point: {}".format(self.best_fit))
# Posterior mean
self.posterior_mean = stats.posterior_mean(*self.pdf_data)
self.summary.append("Posterior mean: {}".format(self.posterior_mean))
# Posterior median
self.posterior_median = one_dim.posterior_median(*self.pdf_data)
self.summary.append("Posterior median: {}".format(
self.posterior_median))
# Posterior mode
self.posterior_modes = one_dim.posterior_mode(*self.pdf_data)
self.summary.append("Posterior mode/s: {}".format(
self.posterior_modes))
def __init__(self, data, plot_options):
super(TwoDimPlot, self).__init__(data, plot_options)
opt = self.plot_options
# If the user didn't specify bin or plot limits,
# we find the extent of the data and use that to set them.
extent = np.zeros(4)
extent[0] = min(self.xdata)
extent[1] = max(self.xdata)
extent[2] = min(self.ydata)
extent[3] = max(self.ydata)
if self.plot_options.bin_limits is None:
self.plot_options = self.plot_options._replace(bin_limits=[[extent[0], extent[1]], [extent[2], extent[3]]])
if self.plot_options.plot_limits is None:
self.plot_options = self.plot_options._replace(plot_limits=extent)
# Posterior PDF
if opt.kde_pdf:
# KDE estimate of PDF
self.pdf_data = two_dim.kde_posterior_pdf(
self.xdata, self.ydata, self.posterior, bw_method=opt.bw_method, bin_limits=opt.bin_limits
)
else:
# Binned estimate of PDF
self.pdf_data = two_dim.posterior_pdf(
self.xdata, self.ydata, self.posterior, nbins=opt.nbins, bin_limits=opt.bin_limits
)
# Profile likelihood
self.prof_data = two_dim.profile_like(
self.xdata, self.ydata, self.chisq, nbins=opt.nbins, bin_limits=opt.bin_limits
)
# As with the 1D plots we use raw data for the best-fit point,
# and binned data for the mean and mode.
# Best-fit point
self.best_fit_x = stats.best_fit(self.chisq, self.xdata)
self.best_fit_y = stats.best_fit(self.chisq, self.ydata)
self.summary.append("Best-fit point (x,y): {}, {}".format(self.best_fit_x, self.best_fit_y))
# Posterior mean
self.posterior_mean_x = stats.posterior_mean(np.sum(self.pdf_data.pdf, axis=1), self.pdf_data.bin_centers_x)
self.posterior_mean_y = stats.posterior_mean(np.sum(self.pdf_data.pdf, axis=0), self.pdf_data.bin_centers_y)
self.summary.append("Posterior mean (x,y): {}, {}".format(self.posterior_mean_x, self.posterior_mean_y))
# Posterior mode
self.posterior_modes = two_dim.posterior_mode(*self.pdf_data)
self.summary.append("Posterior modes/s (x,y): {}".format(self.posterior_modes))
# Posterior median
self.posterior_median_x = one_dim.posterior_median(
np.sum(self.pdf_data.pdf, axis=1), self.pdf_data.bin_centers_x
)
self.posterior_median_y = one_dim.posterior_median(
np.sum(self.pdf_data.pdf, axis=0), self.pdf_data.bin_centers_y
)
self.summary.append("Posterior median (x,y): {}, {}".format(self.posterior_median_x, self.posterior_median_y))
def __init__(self, data, plot_options):
super(OneDimPlot, self).__init__(data, plot_options)
opt = self.plot_options
# If the user didn't specify bin or plot limits,
# we find the extent of the data and use that to set them.
extent = np.zeros(4)
extent[0] = min(self.xdata)
extent[1] = max(self.xdata)
extent[2] = 0
extent[3] = 1.2
# Downside of using named tuple is they're immutable
# so changing options is (justifiably) annoying.
# If this happens a lot (it shouldn't), consider
# using a mutable type instead...
if self.plot_options.bin_limits is None:
self.plot_options = self.plot_options._replace(bin_limits=[extent[0], extent[1]])
if self.plot_options.plot_limits is None:
self.plot_options = self.plot_options._replace(plot_limits=extent)
# Posterior PDF. Norm by area if not showing profile likelihood,
# otherwise norm max value to one.
if opt.kde_pdf:
# KDE estimate of PDF
self.pdf_data = one_dim.kde_posterior_pdf(
self.xdata,
self.posterior,
bin_limits=opt.bin_limits,
norm_area=not opt.show_prof_like,
bw_method=opt.bw_method,
)
else:
# Binned estimate of PDF
self.pdf_data = one_dim.posterior_pdf(
self.xdata, self.posterior, nbins=opt.nbins, bin_limits=opt.bin_limits, norm_area=not opt.show_prof_like
)
# Profile likelihood
self.prof_data = one_dim.prof_data(self.xdata, self.chisq, nbins=opt.nbins, bin_limits=opt.bin_limits)
# Note the best-fit point is calculated using the raw data,
# while the mean, median and mode use the binned PDF.
# Best-fit point
self.best_fit = stats.best_fit(self.chisq, self.xdata)
self.summary.append("Best-fit point: {}".format(self.best_fit))
# Posterior mean
self.posterior_mean = stats.posterior_mean(*self.pdf_data)
self.summary.append("Posterior mean: {}".format(self.posterior_mean))
# Posterior median
self.posterior_median = one_dim.posterior_median(*self.pdf_data)
self.summary.append("Posterior median: {}".format(self.posterior_median))
# Posterior mode
self.posterior_modes = one_dim.posterior_mode(*self.pdf_data)
self.summary.append("Posterior mode/s: {}".format(self.posterior_modes))