def corner(**kwargs):
"""
Evaluate mcmc corner plot.
"""
flat_chain = kwargs.pop('flat_chain')
fit_params = kwargs.pop('fit_params')
variable_labels = kwargs.pop('variable_labels')
figure = _corner(flat_chain, **kwargs)
# Extract the axes
ndim = flat_chain.shape[1]
axes = np.array(figure.axes).reshape((ndim, ndim))
# Loop over the diagonal, adding units
for i, label in enumerate(variable_labels):
ax = axes[i, i]
value = fit_params[label]['value']
bottom = fit_params[label]["confidence_interval"]['min'] - value
top = fit_params[label]["confidence_interval"]['max'] - value
unit = fit_params[label]['unit']
unit = '' if unit == 'dimensionless' or unit is None else unit
title = r'{0}=${1:.2f}^{{{2:+.2f}}}_{{{3:+.2f}}}$ {4}'.format(
kwargs['labels'][i], value, top, bottom, unit)
ax.set_title(title)
plt.show()
def corner(*args, **kwargs):
"""
Override `corner.corner` by making some appearance tweaks.
"""
# Get the usual corner plot
figure = _corner(*args, **kwargs)
# Get the axes
ndim = int(np.sqrt(len(figure.axes)))
axes = np.array(figure.axes).reshape((ndim, ndim))
# Smaller tick labels
for ax in axes[1:, 0]:
for tick in ax.yaxis.get_major_ticks():
tick.label.set_fontsize(10)
formatter = matplotlib.ticker.ScalarFormatter(useOffset=False)
ax.yaxis.set_major_formatter(formatter)
ax.set_ylabel(
ax.get_ylabel(), fontsize=kwargs.get("corner_label_size", 16)
)
for ax in axes[-1, :]:
for tick in ax.xaxis.get_major_ticks():
tick.label.set_fontsize(10)
formatter = matplotlib.ticker.ScalarFormatter(useOffset=False)
ax.xaxis.set_major_formatter(formatter)
ax.set_xlabel(
ax.get_xlabel(), fontsize=kwargs.get("corner_label_size", 16)
)
# Pad the axes to always include the truths
truths = kwargs.get("truths", None)
if truths is not None:
for row in range(1, ndim):
for col in range(row):
lo, hi = np.array(axes[row, col].get_xlim())
if truths[col] < lo:
lo = truths[col] - 0.1 * (hi - truths[col])
axes[row, col].set_xlim(lo, hi)
axes[col, col].set_xlim(lo, hi)
elif truths[col] > hi:
hi = truths[col] - 0.1 * (hi - truths[col])
axes[row, col].set_xlim(lo, hi)
axes[col, col].set_xlim(lo, hi)
lo, hi = np.array(axes[row, col].get_ylim())
if truths[row] < lo:
lo = truths[row] - 0.1 * (hi - truths[row])
axes[row, col].set_ylim(lo, hi)
axes[row, row].set_xlim(lo, hi)
elif truths[row] > hi:
hi = truths[row] - 0.1 * (hi - truths[row])
axes[row, col].set_ylim(lo, hi)
axes[row, row].set_xlim(lo, hi)
return figure
def plot_mcmc_results(sampled_mcmc, bins=20, smooth=None, truths=None):
"""
Uses the 'corner' package to make a *sick* corner plot showing the
projections of the parameters from this MCMC run in a multidimensional
space.
sampled_mcmc: The output of a pymc run.
bins: The number of bins in the histogram.
smooth: Higher number = more smooth (for the 2D plots).
truths: List in [x, spec_index, y] order of known parameter values if applicable.
"""
x_results = sampled_mcmc.x.trace()
y_results = sampled_mcmc.y.trace()
spec_index_results = sampled_mcmc.spec_index.trace()
mcmc_all_results = _np.array([x_results, spec_index_results, y_results]).T
corner_plot = _corner(mcmc_all_results, labels=["x", "spectral index", "y"], bins=bins, smooth=smooth, truths=truths)
return corner_plot
def corner(*args, **kwargs):
if _corner is None:
raise ImportError('could not import corner')
return _corner(*args, **kwargs)