import math
import sys
import arviz as az
import numpy as np
import pandas as pd
import seaborn as sns
from matplotlib import gridspec
from matplotlib import pyplot as plt
from scipy.ndimage import gaussian_filter as gaussian
from scipy.stats import gaussian_kde
from stratmc.data import accumulation_rate, clean_data
from stratmc.inference import (
age_range_to_height,
calculate_lengthscale_stability,
calculate_proxy_signal_stability,
count_samples,
find_gaps,
map_ages_to_section,
)
pd.options.mode.chained_assignment = None
[docs]
def proxy_strat(sample_df, ages_df, proxy = 'd13c', plot_constraints = True, plot_excluded_samples = False, cmap = 'Spectral', legend = False, **kwargs):
"""
Plot stratigraphic data (proxy observations and age constraints) for each section.
.. plot::
from stratmc.data import load_object
from stratmc.plotting import proxy_strat
example_sample_path = 'examples/example_sample_df'
example_ages_path = 'examples/example_ages_df'
sample_df = load_object(example_sample_path)
ages_df = load_object(example_ages_path)
proxy_strat(sample_df, ages_df, proxy = 'd13c')
plt.show()
Parameters
----------
sample_df: pandas.DataFrame
:class:`pandas.DataFrame` containing proxy data for all sections.
ages_df: pandas.DataFrame
:class:`pandas.DataFrame` containing age constraints for all sections.
proxy: str, optional
Name of proxy. Defaults to 'd13c'.
sections: list(str) or numpy.array(str), optional
List of sections to plot. Defaults to all sections in ``sample_df``.
plot_constraints: bool, optional
Whether to plot age constraints as dashed lines. Ages are printed above dashed lines by defalut; to turn off age labels, pass ``print_ages = False``.
plot_excluded_samples: bool, optional
Whether to plot proxy observations for excluded samples (``Exclude?`` is ``True`` in ``sample_df``) as red dots. Defaults to ``False``.
cmap: str, optional
Name of seaborn color palette to use for sections. Defaults to 'Spectral'.
legend: bool, optional
Generate a legend. Defaults to ``False``.
print_ages: bool, optional
Print ages as text on figure.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with observations for each section.
"""
if 'sections' in kwargs:
sections = list(kwargs['sections'])
else:
sections = np.unique(sample_df[~np.isnan(sample_df[proxy])]['section'])
if 'print_ages' in kwargs:
print_ages = kwargs['print_ages']
else:
print_ages = True
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
cs = {}
pal = sns.color_palette(cmap,n_colors=len(sections))
cols = 4
N = len(sections)
rows = int(math.ceil(N/cols))
gs = gridspec.GridSpec(rows, cols)
fig = plt.figure(figsize = (10, 4.5*rows))
for n in range(N):
section = sections[n]
cs[section] = pal[n]
ax = fig.add_subplot(gs[n])
if len(sample_df[proxy][(sample_df['section'] == section) & (sample_df['Exclude?'] == True)]) > 0:
if plot_excluded_samples:
ax.scatter(sample_df[proxy][(sample_df['section'] == section) & (sample_df['Exclude?'] == True)],
sample_df['height'][(sample_df['section'] == section) & (sample_df['Exclude?'] == True)],
color = 'red',
edgecolors = 'k',
label = 'Excluded sample')
ax.scatter(sample_df[proxy][(sample_df['section'] == section) & (sample_df['Exclude?'] != True)],
sample_df['height'][(sample_df['section'] == section) & (sample_df['Exclude?'] != True)],
color = cs[section],
edgecolors = 'k',
label = 'Sample')
xl = ax.get_xlim()
yl = ax.get_ylim()
if plot_constraints:
age_heights = ages_df['height'][(ages_df['section']==section) & (~ages_df['Exclude?']) & (~ages_df['intermediate detrital?']) & (~ages_df['intermediate intrusive?']) & (~ages_df['depositional?'])]
for h in age_heights:
ax.axhline(h,
color = cs[section],
linestyle = 'dashed',
label = 'Depositional age',
zorder = 2)
detrital_age_heights = ages_df['height'][(ages_df['section']==section) & (ages_df['Exclude?'] == False) & (ages_df['intermediate detrital?'] == True)]
for h in detrital_age_heights:
ax.axhline(h,
color = cs[section],
linestyle = 'dotted',
label = 'Detrital age',
zorder = 2)
intrusive_age_heights = ages_df['height'][(ages_df['section']==section) & (ages_df['Exclude?'] == False) & (ages_df['intermediate intrusive?'] == True)]
for h in intrusive_age_heights:
ax.axhline(h,
color = cs[section],
linestyle = 'dashdot',
label = 'Intrusive age',
zorder = 2)
if legend:
ax.legend()
x_range = xl[1] - xl[0]
y_range = yl[1] - yl[0]
if print_ages:
for age, height, sigma, dist_type, param1, param2 in zip(ages_df['age'][(ages_df['section']==section) & (~ages_df['Exclude?']) & (~ages_df['depositional?'])],
ages_df['height'][(ages_df['section']==section) & (ages_df['Exclude?'] != True) & (~ages_df['depositional?'])],
ages_df['age_std'][(ages_df['section']==section) & (ages_df['Exclude?'] != True) & (~ages_df['depositional?'])],
ages_df['distribution_type'][(ages_df['section']==section) & (ages_df['Exclude?'] != True) & (~ages_df['depositional?'])],
ages_df['param_1'][(ages_df['section']==section) & (ages_df['Exclude?'] != True) & (~ages_df['depositional?'])],
ages_df['param_2'][(ages_df['section']==section) & (ages_df['Exclude?'] != True) & (~ages_df['depositional?'])]):
if dist_type == 'Normal':
ax.text(xl[0] + x_range * 0.05, # 0.45
height + y_range * 0.01,
s = f'{age:.1f}' + r'$\pm$' + f'{2*sigma:.1f} Ma',
fontsize = 10)
elif dist_type == 'Uniform':
ax.text(xl[0] + x_range * 0.05, # 0.45
height + y_range * 0.01,
s = f'{param1:.1f} - {param2:.1f} Ma',
fontsize = 10)
ax.set_xlim(xl)
ax.set_ylabel('Height (m)', fontsize = fs)
if proxy == 'd13c':
ax.set_xlabel(r'$\delta^{13}$C$_{\mathrm{carb}} (‰)$', fontsize = fs)
else:
ax.set_xlabel(proxy, fontsize = fs)
ax.set_title(section, fontsize = fs)
ax.tick_params(bottom = True, top = False, left = True, right = False, direction = 'out', labelsize = fs, width = 1.5)
ax.locator_params(axis='x', nbins = 5)
fig.tight_layout()
plt.subplots_adjust(wspace=0.45)
return fig
[docs]
def proxy_inference(sample_df, ages_df, full_trace, legend = True, plot_constraints = False, plot_samples = False, plot_excluded_samples = False, plot_mean = False, plot_mle = False, orientation = 'horizontal', marker_size = 20, section_legend = False, section_cmap = 'Spectral', **kwargs):
"""
Plot the inferred proxy signal over time.
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import proxy_inference
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
example_ages_path = 'examples/example_ages_df'
sample_df = load_object(example_sample_path)
ages_df = load_object(example_ages_path)
proxy_inference(sample_df, ages_df, full_trace, proxy = 'd13c')
plt.show()
Parameters
----------
sample_df: pandas.DataFrame
:class:`pandas.DataFrame` containing proxy data for all sections.
ages_df: pandas.DataFrame
:class:`pandas.DataFrame` containing age constraints for all sections.
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
proxy: str, optional
Tracer to plot; only required if more than one proxy was included in the inference.
legend: bool, optional
Generate a legend. Defaults to ``True``.
plot_constraints: bool, optional
Plot age constraints for each section as dashed lines. Defaults to ``False``.
plot_samples: bool, optional
Plot proxy observations by most likely posterior age. Defaults to ``False``.
plot_excluded_samples: bool, optional
Plot proxy observations that were excluded from the inference (``Exclude?`` is ``True`` in ``sample_df``). Defaults to ``False``.
plot_mean: bool, optional
Plot the mean as a dashed line. Defaults to ``False``.
plot_mle: bool, optional
Plot the maximum likelihood estimate. Defaults to ``False``.
orientation: str, optional
Orientation of figure ('horizontal' with age on the x-axis, or 'vertical' with age on the y-axis). Defaults to 'horizontal'.
marker_size: int, optional
Size of markers if ``plot_samples`` is ``True``. Defaults to 20.
section_legend: bool, optional
Include section names in the legend (if ``plot_samples`` is ``True``). Defaults to ``False``.
section_cmap: str, optional
Name of seaborn color palette to use for sections (if ``plot_samples`` is ``True``). Defaults to 'Spectral'.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with the proxy signal inference.
"""
if orientation == 'horizontal':
horizontal = True
vertical = False
if orientation == 'vertical':
horizontal = False
vertical = True
if 'figsize' in kwargs:
figsize = kwargs['figsize']
elif horizontal:
figsize = (9, 5)
elif vertical:
figsize = (5, 9)
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'proxy' in kwargs:
proxy = kwargs['proxy']
else:
proxy = proxies[0]
if 'sections' in kwargs:
sections = list(kwargs['sections'])
else:
sections = np.unique(sample_df[~np.isnan(sample_df[proxy])]['section'])
sample_df, ages_df = clean_data(sample_df, ages_df, proxies, sections)
ages = full_trace.X_new.X_new.values
fig=plt.figure(figsize=figsize)
ax = fig.gca()
cs = {}
pal = sns.color_palette(section_cmap,n_colors=len(sections))
for i in range(len(sections)):
section = sections[i]
cs[section] = pal[i]
min_ages = []
max_ages = []
if plot_samples: # by max likelihood
plotted_excluded = False
plotted_data = False
for section in sections:
section_df = sample_df[sample_df['section']==section]
proxy_idx = (~np.isnan(section_df[proxy]))
excluded_idx = (~np.isnan(section_df[proxy])) & (section_df['Exclude?'])
sec_ages = az.extract(full_trace.posterior)[str(section)+'_ages'].values
max_like = np.zeros(sec_ages.shape[0])
for i in np.arange(sec_ages.shape[0]):
sample_ages = sec_ages[i,:]
dx = np.linspace(np.min(sample_ages), np.max(sample_ages), 1000)
max_like[i] = dx[np.argmax(gaussian_kde(sample_ages, bw_method = 1)(dx))]
if section_legend:
label = section
elif not plotted_data:
label = 'Sample (most likely age)'
else:
label = '_nolegend_'
if vertical:
ax.scatter(section_df[proxy][proxy_idx & ~excluded_idx],
max_like[proxy_idx & ~excluded_idx],
#np.mean(sec_ages, axis = 1),
s = marker_size,
color = cs[section],
label = label,
edgecolors = 'k',
lw = 0.5,
zorder = 5)
plotted_data = True
if (plot_excluded_samples) and (len(section_df[proxy][excluded_idx]) > 0):
if not plotted_excluded:
excluded_label = 'Excluded samples'
else:
excluded_label = '_nolegend'
ax.scatter(section_df[proxy][excluded_idx & proxy_idx],
max_like[excluded_idx & proxy_idx],
s = marker_size,
color = 'none',
label = excluded_label,
edgecolors = cs[section],
lw = 0.5,
zorder = 5)
plotted_excluded = True
if horizontal:
ax.scatter(max_like[proxy_idx & ~excluded_idx],
section_df[proxy][proxy_idx & ~excluded_idx],
s = marker_size,
color = cs[section],
label = label,
lw = 0.5,
edgecolors = 'k',
zorder = 5)
plotted_data = True
if (plot_excluded_samples) and (len(section_df[proxy][proxy_idx & excluded_idx]) > 0):
if not plotted_excluded:
excluded_label = 'Excluded samples'
else:
excluded_label = '_nolegend'
ax.scatter(max_like[excluded_idx & proxy_idx],
section_df[proxy][excluded_idx & proxy_idx],
s = marker_size,
color = 'none',
label = excluded_label,
edgecolors = cs[section],
lw = 0.5,
zorder = 5)
plotted_excluded = True
min_ages = np.append(min_ages, np.min(np.mean(sec_ages,axis = 1)))
max_ages = np.append(max_ages, np.max(np.mean(sec_ages, axis = 1)))
proxy_pred = az.extract(full_trace.posterior_predictive)['f_pred_' + proxy].values
lo = np.percentile(proxy_pred, 2.5, axis=1).flatten()
hi = np.percentile(proxy_pred, 97.5, axis=1).flatten()
if vertical:
ax.fill_betweenx(ages.ravel(),
hi,
lo,
color=(.95,.95,.95),
label='95% envelope',
linestyle = 'dotted',
lw = 1.5,
edgecolor = 'k',
zorder = 0)
if horizontal:
ax.fill_between(ages.ravel(),
hi,
lo,
color=(.95,.95,.95),
label='95% envelope',
linestyle = 'dotted',
edgecolor = 'k',
lw = 1.5,
zorder = 0)
lo = np.percentile(proxy_pred, 17, axis=1).flatten()
hi = np.percentile(proxy_pred, 100-17, axis=1).flatten()
if vertical:
ax.fill_betweenx(ages.ravel(),
hi,
lo,
color=(.8,.8,.8),
linestyle = 'dashed',
label='66% envelope',
edgecolor = 'k',
lw = 1.5,
zorder = 1)
if horizontal:
ax.fill_between(ages.ravel(),
hi,
lo,
color=(.8,.8,.8),
linestyle = 'dashed',
label='66% envelope',
edgecolor = 'k',
lw = 1.5,
zorder = 1)
lo = np.percentile(proxy_pred, 33.5, axis=1).flatten()
hi = np.percentile(proxy_pred, 100-33.5, axis=1).flatten()
if vertical:
ax.fill_betweenx(ages.ravel(),
hi,
lo,
color=(.65,.65,.65),
label='33% envelope',
edgecolor = 'k',
lw = 1.5,
zorder = 2)
if horizontal:
ax.fill_between(ages.ravel(),
hi,
lo,
color=(.65,.65,.65),
label='33% envelope',
edgecolor = 'k',
lw = 1.5,
zorder = 2)
if plot_mean:
if horizontal:
ax.plot(ages.ravel(),
np.mean(proxy_pred,axis=1).flatten(),
color = 'k',
linestyle = 'solid',
lw = 2,
zorder = 3,
label = 'Mean ' + str(proxy))
if vertical:
ax.plot(np.mean(proxy_pred, axis=1).flatten(),
ages.ravel(),
color = 'k',
linestyle = 'solid',
lw = 2,
zorder = 3,
label = 'Mean ' + str(proxy))
if plot_mle:
dy = np.linspace(np.min(proxy_pred), np.max(proxy_pred), 200)
max_like = np.zeros(ages.size)
for i in np.arange(ages.size):
time_slice = proxy_pred[i,:]
max_like[i] = dy[np.argmax(gaussian_kde(time_slice, bw_method = 1)(dy))]
max_like = gaussian(max_like, 2)
if horizontal:
ax.plot(ages.ravel(),
max_like,
zorder = 4,
color = 'k',
linestyle = 'solid',
lw = 2,
label = 'Most likely ' + str(proxy))
if vertical:
ax.plot(max_like,
ages.ravel(),
zorder = 4,
color = 'k',
linestyle = 'solid',
lw = 2,
label = 'Most likely ' + str(proxy))
xl = ax.get_xlim()
yl = ax.get_ylim()
if plot_constraints:
i = 0
for section in sections:
if i == 0:
label = 'Age constraint'
else:
label = '_nolegend_'
if vertical:
ax.hlines(ages_df['age'][ages_df['section']==section],
xl[0],
xl[1],
color = cs[section],
linestyle = 'dashed',
label = label,
zorder = -1)
ax.set_xlim(xl)
if horizontal:
ax.vlines(ages_df['age'][ages_df['section']==section],
yl[0],
yl[1],
color = cs[section],
linestyle = 'dashed',
label = label,
zorder = -1)
ax.set_ylim(yl)
i += 1
min_ages = np.append(min_ages, np.min(ages))
max_ages = np.append(max_ages, np.max(ages))
if vertical:
ax.set_ylabel('Age (Ma)', fontsize = fs)
if proxy == 'd13c':
ax.set_xlabel(r'$\delta^{13}$C$_{\mathrm{carb}}$ (‰)', fontsize = fs)
else:
ax.set_xlabel(proxy, fontsize = fs)
ax.set_ylim([np.min(min_ages) - 1, np.max(max_ages) + 1])
ax.invert_yaxis()
if horizontal:
ax.set_xlabel('Age (Ma)', fontsize = fs)
if proxy == 'd13c':
ax.set_ylabel(r'$\delta^{13}$C$_{\mathrm{carb}}$ (‰)', fontsize = fs)
else:
ax.set_ylabel(proxy, fontsize = fs)
ax.set_xlim([np.min(min_ages) - 1, np.max(max_ages) + 1])
ax.invert_xaxis()
ax.tick_params(bottom = True, top = True, left = True, right = True, direction = 'in', labelsize = fs, width = 1.5)
# for axis in ['top','bottom','left','right']:
# ax.spines[axis].set_linewidth(1.5)
if legend:
ax.legend(loc='best')
return fig
[docs]
def interpolated_proxy_inference(interpolated_df, interpolated_proxy_df, proxy, legend = True, plot_samples = False, plot_mle = True, orientation = 'horizontal', section_legend = False, marker_size = 20, section_cmap = 'Spectral', **kwargs):
"""
Plot interpolated proxy signal over time (by extending the posterior section age models to a new proxy not included in the inference model) using interpolated age models from :py:meth:`extend_age_model() <stratmc.inference.extend_age_model>` and interpolated proxy values from :py:meth:`interpolate_proxy() <stratmc.inference.interpolate_proxy>` in :py:mod:`stratmc.inference`.
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.inference import extend_age_model, interpolate_proxy
from stratmc.plotting import interpolated_proxy_inference
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
example_sample_path_d18o = 'examples/example_sample_df_d18o'
example_ages_path = 'examples/example_ages_df'
sample_df = load_object(example_sample_path)
sample_df_d18o = load_object(example_sample_path_d18o)
ages_df = load_object(example_ages_path)
interpolated_df = extend_age_model(full_trace, sample_df, ages_df, ['d18o'], new_proxy_df = sample_df_d18o)
ages_new = full_trace.X_new.X_new.values.ravel()
interpolated_proxy_df = interpolate_proxy(interpolated_df, 'd18o', ages_new)
interpolated_proxy_inference(interpolated_df, interpolated_proxy_df, 'd18o')
plt.show()
Parameters
----------
interpolated_df: pandas.DataFrame
:class:`pandas.DataFrame` with interpolated age draws and sample age summary statistics from :py:meth:`extend_age_model() <stratmc.inference.extend_age_model>` in :py:mod:`stratmc.inference`.
interpolated_proxy_df: pandas.DataFrame
:class:`pandas.DataFrame` with interpolated proxy values and summary statistics at target ages from :py:meth:`interpolate_proxy() <stratmc.inference.interpolate_proxy>` in :py:mod:`stratmc.inference`.
proxy: str
Name of new proxy (must match column name in ``interpolated_proxy_df``).
legend: bool, optional
Generate a legend. Defaults to ``True``.
plot_samples: bool, optional
Plot proxy observations by most likely posterior age. Defaults to ``False``.
plot_mle: bool, optional
Plot the maximum likelihood estimate. Defaults to ``True``.
orientation: str, optional
Orientation of figure ('horizontal' or 'vertical'). Defaults to 'horizontal'.
marker_size: int, optional
Size of markers if ``plot_samples`` is ``True``. Defaults to 20.
section_legend: bool, optional
Include section names in the legend (if ``plot_samples`` is ``True``). Defaults to ``False``.
section_cmap: str, optional
Name of seaborn color palette to use for sections (if ``plot_samples`` is ``True``). Defaults to 'Spectral'.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with interpolated proxy signal over time.
"""
if 'sections' in kwargs:
sections = list(kwargs['sections'])
else:
sections = np.unique(interpolated_df['section'])
if orientation == 'horizontal':
horizontal = True
vertical = False
if orientation == 'vertical':
horizontal = False
vertical = True
if 'figsize' in kwargs:
figsize = kwargs['figsize']
elif horizontal:
figsize = (9, 5)
elif vertical:
figsize = (5, 9)
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
fig=plt.figure(figsize=figsize)
ax = fig.gca()
cs = {}
pal = sns.color_palette(section_cmap,n_colors=len(sections))
for i in range(len(sections)):
section = sections[i]
cs[section] = pal[i]
min_ages = []
max_ages = []
if plot_samples: # by max likelihood
for section in sections:
section_df = interpolated_df[interpolated_df['section']==section]
if section_legend:
label = section
else:
label = '_nolegend_'
if vertical:
ax.scatter(section_df[proxy],
section_df['mle'],
s = marker_size,
color = cs[section],
label = label,
edgecolors = 'k',
lw = 0.5,
zorder = 3)
if horizontal:
ax.scatter(section_df['mle'],
section_df[proxy],
s = marker_size,
color = cs[section],
label = label,
lw = 0.5,
edgecolors = 'k',
zorder = 3)
sec_ages = section_df['mle'].values
min_ages = np.append(min_ages, np.min(sec_ages))
max_ages = np.append(max_ages, np.max(sec_ages))
if vertical:
ax.fill_betweenx(interpolated_proxy_df['age'],
interpolated_proxy_df['97.5'],
interpolated_proxy_df['2.5'],
color=(.95,.95,.95),
label='95% envelope',
linestyle = '--',
lw = 1.5,
edgecolor = 'k',
zorder = 0)
if horizontal:
ax.fill_between(interpolated_proxy_df['age'],
interpolated_proxy_df['97.5'],
interpolated_proxy_df['2.5'],
color=(.95,.95,.95),
label='95% envelope',
linestyle = '--',
edgecolor = 'k',
lw = 1.5,
zorder = 0)
if vertical:
ax.fill_betweenx(interpolated_proxy_df['age'],
interpolated_proxy_df['84'],
interpolated_proxy_df['16'],
color=(.8,.8,.8),
label='68% envelope',
edgecolor = 'k',
lw = 1.5,
zorder = 1)
if horizontal:
ax.fill_between(interpolated_proxy_df['age'],
interpolated_proxy_df['84'],
interpolated_proxy_df['16'],
color=(.8,.8,.8),
label='68% envelope',
edgecolor = 'k',
lw = 1.5,
zorder = 1)
if plot_mle:
if horizontal:
ax.plot(interpolated_proxy_df['age'], interpolated_proxy_df['mle'], zorder = 2, color = 'k', linestyle = 'solid', lw = 2, label = 'Most likely ' + str(proxy))
if vertical:
ax.plot(interpolated_proxy_df['mle'], interpolated_proxy_df['age'], zorder = 2, color = 'k', linestyle = 'solid', lw = 2, label = 'Most likely ' + str(proxy))
min_ages = np.append(min_ages, np.min(interpolated_proxy_df['age']))
max_ages = np.append(max_ages, np.max(interpolated_proxy_df['age']))
if vertical:
ax.set_ylabel('Age (Ma)', fontsize = fs)
ax.set_xlabel(proxy, fontsize = fs)
ax.set_ylim([np.min(min_ages) - 1, np.max(max_ages) + 1])
ax.invert_yaxis()
if horizontal:
ax.set_xlabel('Age (Ma)', fontsize = fs)
ax.set_ylabel(proxy, fontsize = fs)
ax.set_xlim([np.min(min_ages) - 1, np.max(max_ages) + 1])
ax.invert_xaxis()
ax.tick_params(bottom = True, top = True, left = True, right = True, direction = 'in', labelsize = fs, width = 1.5)
if legend:
ax.legend(loc='best')
return fig
[docs]
def age_height_model(sample_df, ages_df, full_trace, include_excluded_samples = True, plot_samples = True, cmap = 'Spectral', legend = False, **kwargs):
"""
Generate a posterior age-height plot for each section.
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import age_height_model
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
example_ages_path = 'examples/example_ages_df'
sample_df = load_object(example_sample_path)
ages_df = load_object(example_ages_path)
age_height_model(sample_df, ages_df, full_trace)
plt.show()
Parameters
----------
sample_df: pandas.DataFrame
:class:`pandas.DataFrame` containing proxy data for all sections.
ages_df: pandas.DataFrame
:class:`pandas.DataFrame` containing age constraints for all sections.
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
sections: list(str) or numpy.array(str), optional
List of sections to plot. Defaults to all sections in ``sample_df``.
cmap: str, optional
Name of seaborn color palette to use for sections. Defaults to 'Spectral'.
legend: bool, optional
Generate a legend. Defaults to ``False``.
include_excluded_samples: bool, optional
Whether to consider excluded samples (``Exclude?`` is ``True`` in ``sample_df``) whose ages were passively tracked in the inference model. Defaults to ``True``.
plot_samples: bool, optional
Plot proxy observations by most likely posterior age. Defaults to ``True``.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with age-height models for each section.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'sections' in kwargs:
sections = list(kwargs['sections'])
else:
sections = np.unique(sample_df.dropna(subset = proxies, how = 'all')['section'])
sample_df, ages_df = clean_data(sample_df, ages_df, proxies, sections)
cs = {}
pal = sns.color_palette(cmap, n_colors=len(sections))
cols = 4
N = len(sections)
rows = int(math.ceil(N/cols))
gs = gridspec.GridSpec(rows, cols)
fig = plt.figure(figsize = (11, 2.5*rows))
for n in range(N):
section = sections[n]
cs[section] = pal[n]
ax = fig.add_subplot(gs[n])
sec_ages = az.extract(full_trace.posterior)[str(section)+'_ages'].values # posterior_predictive
section_df = sample_df[sample_df['section']==section]
lo = np.percentile(sec_ages, 2.5, axis=1).flatten()
hi = np.percentile(sec_ages, 97.5, axis=1).flatten()
if include_excluded_samples:
ax.fill_betweenx(section_df['height'],
hi,
lo,
color=(.95,.95,.95),
linestyle = 'dotted',
label=r'95% envelope')
else:
included_idx = ~section_df['Exclude?'].values.astype(bool)
ax.fill_betweenx(section_df['height'][included_idx],
hi[included_idx],
lo[included_idx],
color=(.95,.95,.95),
linestyle='dotted',
label=r'95% envelope')
lo = np.percentile(sec_ages,17,axis=1).flatten()
hi = np.percentile(sec_ages,100-17,axis=1).flatten()
if include_excluded_samples:
ax.fill_betweenx(section_df['height'],
hi,
lo,
color=(.8,.8,.8),
linestyle='dashed',
label=r'66% envelope')
else:
ax.fill_betweenx(section_df['height'][included_idx],
hi[included_idx],
lo[included_idx],
color=(.8,.8,.8),
linestyle='dashed',
label=r'66% envelope')
lo = np.percentile(sec_ages,33.5,axis=1).flatten()
hi = np.percentile(sec_ages,100-33.5,axis=1).flatten()
if include_excluded_samples:
ax.fill_betweenx(section_df['height'],
hi,
lo,
color=(.65,.65,.65),
linestyle='solid',
label=r'33% envelope')
else:
ax.fill_betweenx(section_df['height'][included_idx],
hi[included_idx],
lo[included_idx],
color=(.65,.65,.65),
linestyle='solid',
label=r'33% envelope')
if plot_samples:
# calculate MLE
max_like = np.zeros(sec_ages.shape[0])
for i in np.arange(sec_ages.shape[0]):
sample_ages = sec_ages[i,:]
dx = np.linspace(np.min(sample_ages), np.max(sample_ages), 1000)
max_like[i] = dx[np.argmax(gaussian_kde(sample_ages, bw_method = 1)(dx))]
if include_excluded_samples:
ax.scatter(max_like,
section_df['height'],
color = cs[section],
linewidth = 1,
linestyle = 'solid',
marker = 'x',
s = 10,
label = 'Most likely sample age')
else:
ax.scatter(max_like[included_idx],
section_df['height'][included_idx],
color = cs[section],
linewidth = 1,
linestyle = 'solid',
marker = 'x',
s = 10,
label = 'Most likely sample age')
ax.errorbar(ages_df['age'][ages_df['section']==section],
ages_df['height'][ages_df['section']==section],
xerr = 2*ages_df['age_std'][ages_df['section']==section],
color = cs[section], fmt = 'none', capsize = 3)
section_ages_df = ages_df[(ages_df['section']==section) & (~ages_df['Exclude?'])]
ax.scatter(section_ages_df['age'][(~section_ages_df['intermediate intrusive?']) & (~section_ages_df['intermediate detrital?'])],
section_ages_df['height'][(~section_ages_df['intermediate intrusive?']) & (~section_ages_df['intermediate detrital?'])],
color = cs[section],
lw = 0.5,
edgecolor = 'k',
label = r'Age ($\pm2\sigma$)')
if section_ages_df['age'][section_ages_df['intermediate intrusive?']].shape[0] > 0:
ax.scatter(section_ages_df['age'][section_ages_df['intermediate intrusive?']],
section_ages_df['height'][section_ages_df['intermediate intrusive?']],
marker = 's',
color = cs[section],
edgecolor = 'k',
lw = 0.5,
label = r'Intrusive Age ($\pm2\sigma$)')
if section_ages_df['age'][section_ages_df['intermediate detrital?']].shape[0] > 0:
ax.scatter(section_ages_df['age'][section_ages_df['intermediate detrital?']],
section_ages_df['height'][section_ages_df['intermediate detrital?']],
marker = '^',
color = cs[section],
edgecolor = 'k',
lw = 0.5,
label = r'Detrital Age ($\pm2\sigma$)')
if legend:
plt.legend(loc = 'upper right',
bbox_to_anchor=(1, 1),
markerfirst = False,
frameon = False,
fontsize = 8)
ax.axis('tight')
ax.set_xlabel('Age (Ma)', fontsize = fs)
ax.set_ylabel('Height (m)', fontsize = fs)
ax.tick_params(bottom = True, top = True, left = True, right = True, direction = 'in', labelsize = fs)
ax.set_title(section, fontsize = fs)
fig.tight_layout()
plt.subplots_adjust(wspace=0.35)
return fig
[docs]
def section_proxy_signal(full_trace, sample_df, ages_df, include_radiometric_ages = False, plot_constraints = False, plot_mle = False, yax = 'height', legend = False, cmap = 'Spectral', **kwargs):
"""
Map the posterior proxy signal back to height in each section (using its most likely posterior age model), and plot alongside the proxy observations (plotted by most likely posterior age).
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import section_proxy_signal
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
example_ages_path = 'examples/example_ages_df'
sample_df = load_object(example_sample_path)
ages_df = load_object(example_ages_path)
section_proxy_signal(full_trace, sample_df, ages_df)
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
sample_df: pandas.DataFrame
:class:`pandas.DataFrame` containing proxy data for all sections.
ages_df: pandas.DataFrame
:class:`pandas.DataFrame` containing age constraints for all sections.
include_radiometric_ages: bool, optional
Whether to consider radiometric ages in the posterior age model for each section. Defaults to ``False``.
plot_constraints: bool, optional
Plot age constraints for each section as dashed lines. Defaults to ``False``.
plot_mle: bool, optional
Plot the maximum likelihood estimate for the proxy signal as a line. Defaults to ``False``.
yax: str, optional
Scale for the y-axis ('height' or 'age'). Defaults to 'height'.
legend: bool, optional
Generate a legend. Defaults to ``True``.
cmap: str, optional
Name of seaborn color palette to use for sections. Defaults to 'Spectral'.
proxy: str, optional
Tracer to plot; only required if more than one proxy was included in the inference.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with inferred proxy signal mapped to height in each section.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'proxy' in kwargs:
proxy = kwargs['proxy']
else:
proxy = proxies[0]
if 'sections' in kwargs:
sections = list(kwargs['sections'])
else:
# only want to map signal back to sections that actually have data for this proxy
sections = np.unique(sample_df.dropna(subset = [proxy], how = 'all')['section'])
sample_df, ages_df = clean_data(sample_df, ages_df, proxies, sections)
cs = {}
pal = sns.color_palette(cmap, n_colors=len(sections))
N = len(sections)
cols = np.min([N, 4])
rows = int(math.ceil(N/cols))
gs = gridspec.GridSpec(rows, cols)
fig = plt.figure(figsize = (2.5 * cols, 4*rows))
proxy_pred = az.extract(full_trace.posterior_predictive)['f_pred_' + proxy].values
ages_new = full_trace.X_new.X_new.values
if plot_mle:
# calculate MLE
dy = np.linspace(np.min(proxy_pred), np.max(proxy_pred), 200)
max_like = np.zeros(ages_new.size)
for i in np.arange(ages_new.size):
time_slice = proxy_pred[i,:]
max_like[i] = dy[np.argmax(gaussian_kde(time_slice, bw_method = 1)(dy))]
max_like = gaussian(max_like, 2)
mapped_age_models = map_ages_to_section(full_trace, sample_df, ages_df, sections = sections, include_radiometric_ages = include_radiometric_ages)
for n in range(N):
section = sections[n]
cs[section] = pal[n]
ax = fig.add_subplot(gs[n])
ax.tick_params(bottom = True, top = False, left = True, right = False, direction = 'out', labelsize = fs, width = 1.5)
ax.set_title(section, fontsize = fs)
# calculate maximum likelihood age for each sample
section_df = sample_df[sample_df['section']==section]
proxy_idx = (~np.isnan(section_df[proxy])) & (~section_df['Exclude?'].values.astype(bool))
excluded_idx = (~np.isnan(section_df[proxy])) & (section_df['Exclude?'])
section_age_model = mapped_age_models[mapped_age_models['section']==section]
section_age_vec = section_age_model['age'].values
above = (ages_new <= np.max(section_age_vec))
below = (ages_new >= np.min(section_age_vec))
age_idx = above & below
# plot data
if yax == 'height':
# samples included in inference
ax.scatter(section_df[proxy].values[proxy_idx],
section_df['height'].values[proxy_idx],
color = cs[section],
edgecolor = 'k',
zorder = 3)
# excluded samples
ax.scatter(section_df[proxy].values[excluded_idx],
section_df['height'].values[excluded_idx],
color = 'none',
edgecolor = cs[section],
zorder = 4)
# plot posterior proxy signal
hi = np.percentile(proxy_pred, 97.5, axis=1).flatten()
lo = np.percentile(proxy_pred, 2.5, axis=1).flatten()
ax.fill_betweenx(section_age_model['interpolated height'],
hi[age_idx],
lo[age_idx],
color=(.95,.95,.95),
label='95% envelope',
linestyle = 'dotted',
lw = 1.5,
edgecolor = 'k',
zorder = 0)
hi = np.percentile(proxy_pred, 100-17, axis=1).flatten()
lo = np.percentile(proxy_pred, 17, axis=1).flatten()
ax.fill_betweenx(section_age_model['interpolated height'],
hi[age_idx],
lo[age_idx],
color=(.8,.8,.8),
label='66% envelope',
linestyle = 'dashed',
lw = 1.5,
edgecolor = 'k',
zorder = 1)
hi = np.percentile(proxy_pred, 100-33.5, axis=1).flatten()
lo = np.percentile(proxy_pred, 33.5, axis=1).flatten()
ax.fill_betweenx(section_age_model['interpolated height'],
hi[age_idx],
lo[age_idx],
color=(.65,.65,.65),
label='33% envelope',
linestyle = 'solid',
lw = 1.5,
edgecolor = 'k',
zorder = 2)
if plot_mle:
ax.plot(max_like[age_idx],
section_age_model['interpolated height'],
color = 'k',
lw = 2,
label = 'Most likely',
zorder = 3)
if plot_constraints:
section_ages_df = ages_df[(ages_df['section']==section) & (~ages_df['Exclude?'])]
for h in section_ages_df['height']:
ax.axhline(h, color = cs[section], linestyle = 'dashed', zorder = -1)
if proxy == 'd13c':
ax.set_xlabel(r'$\delta^{13}$C$_{\mathrm{carb}}$ (‰)', fontsize = fs)
else:
ax.set_xlabel(proxy, fontsize = fs)
elif yax == 'age':
# plot posterior proxy signal
hi = np.percentile(proxy_pred, 97.5, axis=1).flatten()
lo = np.percentile(proxy_pred, 2.5, axis=1).flatten()
ax.fill_betweenx(ages_new[age_idx],
hi[age_idx],
lo[age_idx],
color=(.95,.95,.95),
label='95% envelope',
linestyle = 'dotted',
lw = 1.5,
edgecolor = 'k',
zorder = 0)
hi = np.percentile(proxy_pred, 100-17, axis=1).flatten()
lo = np.percentile(proxy_pred, 17, axis=1).flatten()
ax.fill_betweenx(ages_new[age_idx],
hi[age_idx],
lo[age_idx],
color=(.8,.8,.8),
label='66% envelope',
linestyle = 'dashed',
lw = 1.5,
edgecolor = 'k',
zorder = 1)
hi = np.percentile(proxy_pred, 100-33.5, axis=1).flatten()
lo = np.percentile(proxy_pred, 33.5, axis=1).flatten()
ax.fill_betweenx(ages_new[age_idx],
hi[age_idx],
lo[age_idx],
color=(.65,.65,.65),
label='33% envelope',
linestyle = 'solid',
lw = 1.5,
edgecolor = 'k',
zorder = 2)
if plot_mle:
ax.plot(max_like[age_idx],
ages_new[age_idx],
color = 'k',
lw = 2,
label = 'Most likely',
zorder = 3)
sec_ages = az.extract(full_trace.posterior)[str(section)+'_ages'].values
max_like_samples = np.zeros(sec_ages.shape[0])
for i in np.arange(sec_ages.shape[0]):
sample_ages = sec_ages[i,:]
dx = np.linspace(np.min(sample_ages), np.max(sample_ages), 1000)
max_like_samples[i] = dx[np.argmax(gaussian_kde(sample_ages, bw_method = 1)(dx))]
ax.scatter(section_df[proxy].values[proxy_idx],
max_like_samples[proxy_idx],
color = cs[section],
edgecolor = 'k',
zorder = 4)
ax.scatter(section_df[proxy].values[excluded_idx],
max_like_samples[excluded_idx],
color = 'none',
edgecolor = cs[section],
zorder = 5)
if plot_constraints:
section_ages_df = ages_df[(ages_df['section']==section) & (~ages_df['Exclude?'])]
for age in section_ages_df['age']:
ax.axhline(age, color = cs[section], linestyle = 'dashed', zorder = -1)
ax.invert_yaxis()
if proxy == 'd13c':
ax.set_xlabel(r'$\delta^{13}$C$_{\mathrm{carb}}$ (‰)', fontsize = fs)
else:
ax.set_xlabel(proxy, fontsize = fs)
if (n == 0) | ((n % 4) == 0):
if yax == 'height':
ax.set_ylabel('Height (m)', fontsize = fs)
elif yax == 'age':
ax.set_ylabel('Age (Ma)', fontsize = fs)
if (n == 0) and legend:
ax.legend()
fig.tight_layout()
return fig
[docs]
def covariance_hyperparameters(full_trace, figsize = (4, 3.5), **kwargs):
"""
Plot prior and posterior distributions for the lengthscale (:math:`\\ell`) and variance (:math:`\\sigma`) hyperparameters of the :class:`pymc.gp.cov.ExpQuad <pymc.gp.cov.ExpQuad>` Gaussian process covariance kernel:
.. math::
k(x, x') = \\sigma^2 \\mathrm{exp} \\left[-\\frac{(x - x')^2}{2 \\ell^2} \\right]
.. plot::
from stratmc.data import load_trace
from stratmc.plotting import covariance_hyperparameters
full_trace = load_trace('examples/example_docs_trace')
covariance_hyperparameters(full_trace)
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
proxy: str, optional
Tracer to plot model parameters for (each proxy has a different covariance kernel); only required if more than one proxy was included in the inference.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with prior and posterior model parameters.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'proxy' in kwargs:
proxy = kwargs['proxy']
else:
proxy = proxies[0]
fig, ax = plt.subplots(2, figsize = figsize)
sns.kdeplot(full_trace.prior['gp_ls_' + proxy].values.ravel(),
color='indianred',
label = 'Prior',
ax = ax[0],
cut = 0)
sns.kdeplot(full_trace.posterior['gp_ls_' + proxy].values.ravel(),
color = 'indianred',
label = 'Posterior',
fill = True,
edgecolor = 'None',
ax = ax[0],
cut = 0)
ax[0].set_xlabel('Lengthscale', fontsize=fs)
ax[0].ticklabel_format(axis="x", style="sci", scilimits=(0,0))
ax[0].tick_params(labelsize=fs)
ax[0].set_ylabel("")
ax[0].set_yticks([])
ax[0].ticklabel_format(style='plain')
ax[0].legend()
sns.kdeplot(full_trace.prior['gp_var_' + proxy].values.ravel(),
color='#87BED5',
label = 'prior',
ax = ax[1],
cut = 0)
sns.kdeplot(full_trace.posterior['gp_var_' + proxy].values.ravel(),
color = '#87BED5',
label = 'posterior',
fill = True,
edgecolor = 'None',
ax = ax[1],
cut = 0)
ax[1].set_xlabel('Variance', fontsize=fs)
ax[1].ticklabel_format(axis="x", style="sci", scilimits=(0,0))
ax[1].tick_params(labelsize=fs)
ax[1].set_ylabel("")
ax[1].set_yticks([])
ax[1].ticklabel_format(style='plain')
ax[1].legend()
fig.tight_layout()
return fig
[docs]
def section_summary(sample_df, ages_df, full_trace, section, plot_excluded_samples = False, plot_noise_prior = False, plot_offset_prior = False, include_age_constraints_sedrate = True, figsize = (8, 9)):
"""
For a given section, plot posterior estimates of sample age, sedimentation rate, noise, and offset. Noise and offset terms must be either per-section or global; to plot per-sample noise and offset terms, use :py:meth:`noise_summary() <stratmc.plotting>` and :py:meth:`offset_summary() <stratmc.plotting>`.
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import section_summary
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
example_ages_path = 'examples/example_ages_df'
sample_df = load_object(example_sample_path)
ages_df = load_object(example_ages_path)
section_summary(sample_df, ages_df, full_trace, '1')
plt.show()
Parameters
----------
sample_df: pandas.DataFrame
:class:`pandas.DataFrame` containing proxy data for all sections.
ages_df: pandas.DataFrame
:class:`pandas.DataFrame` containing age constraints for all sections.
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
section: str
Name of target section.
plot_excluded_samples: bool, optional
Plot age estimates for proxy observations that were excluded from the inference (``Exclude?`` is ``True`` in ``sample_df``). Defaults to ``False``.
plot_noise_prior: bool, optional
Plot prior distribution for noise term. Defaults to ``False``.
plot_offset_prior: bool, optional
Plot prior distribution for offset term. Defaults to ``False``.
include_age_constraints_sedrate: bool, optional
Include age constraints in sedimentation rate calculations. Defaults to ``True``.
Returns
-------
fig: matplotlib.pyplot.figure
Figure summarizing posterior sample ages, sedimentation rate, and posterior noise and offset terms for the input section.
"""
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if str(section) + '_section_noise_' + proxies[0] in list(full_trace.posterior.keys()):
noise_type = 'section'
else:
noise_type = 'groups'
if str(section) + '_section_offset_' + proxies[0] in list(full_trace.posterior.keys()):
offset_type = 'section'
else:
group_offset_vars = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if f"{'_group_offset_'}" in l
]
if len(group_offset_vars) > 0:
offset_type = 'groups'
else:
offset_type = 'none'
if (noise_type == 'none') and (offset_type == 'none'):
n_rows = 3
elif ((noise_type != 'none') or (offset_type != 'none')) and not ((noise_type != 'none') and (offset_type != 'none')):
n_rows = 4
elif ((noise_type != 'none') and (offset_type != 'none')):
n_rows = 5
sample_df, ages_df = clean_data(sample_df, ages_df, proxies, [section])
fig, ax = plt.subplots(n_rows, 1, figsize = figsize, constrained_layout=True, gridspec_kw={'height_ratios': [1, 1, 1.5] + [0.5] * (n_rows - 3)})
# link axes with age on the x-axis
ax[0].sharex(ax[1])
ax[2].sharex(ax[1])
pal = sns.color_palette("flare", n_colors = np.mean(az.extract(full_trace.posterior)[str(section)+'_ages'].values, axis = 1).size, desat=0.8)
sns.set_palette(pal)
samples = np.mean(az.extract(full_trace.posterior)[str(section)+'_ages'], axis = 1).size # posterior_predictive
sec_ages = az.extract(full_trace.posterior)[str(section)+'_ages'].values # posterior_predictive
section_df = sample_df[sample_df['section']==section]
excluded_idx = section_df['Exclude?'].values
for i in range(samples):
if not excluded_idx[i]:
sns.kdeplot(sec_ages[i,:].ravel().astype(float),
label = 'posterior',
fill = True,
edgecolor = None,
color = pal[i],
alpha = 0.3,
ax = ax[0],
cut = 0)
else:
if plot_excluded_samples:
sns.kdeplot(sec_ages[i,:].ravel().astype(float),
label = 'posterior',
fill = True,
edgecolor = None,
color = pal[i],
alpha = 0.3,
ax = ax[0],
cut = 0)
yl = ax[0].get_ylim()
ax[0].vlines(ages_df['age'][ages_df['section']==section],
yl[0],
yl[1],
color = '#8c8684',
linestyle = 'dashed',
zorder = 3,
label = 'Age constraint')
for i in range(len(ages_df['age'][ages_df['section']==section])):
ax[0].axvspan(ages_df['age'][ages_df['section']==section].values[i] - ages_df['age_std'][ages_df['section']==section].values[i],
ages_df['age'][ages_df['section']==section].values[i] + ages_df['age_std'][ages_df['section']==section].values[i],
alpha=0.5,
color ='#8c8684')
ax[0].invert_xaxis()
ax[0].set_ylim(yl)
ax[0].set_ylabel('Sample ages', fontsize = 12)
ax[0].set_yticks([])
# age constraint posteriors
constraint_ages = az.extract(full_trace.posterior)[str(section)+'_radiometric_age'].values
prior_constraint_ages = az.extract(full_trace.prior)[str(section)+'_radiometric_age'].values
samples = np.mean(constraint_ages, axis = 1).size
pal = sns.color_palette('crest',n_colors=samples)
for i in range(samples):
if i == 0:
post_label = 'Posterior'
prior_label = 'Prior'
else:
post_label = '_nolegend'
prior_label = '_nolegend'
sns.kdeplot(constraint_ages[i,:].ravel(),
label = post_label, fill = True, color = pal[i], edgecolor='None', alpha = 0.3, ax = ax[1], cut = 0)
sns.kdeplot(prior_constraint_ages[i,:].ravel(),
label = prior_label, color = pal[i], fill = False, ax = ax[1], cut = 0)
ax[1].set_ylabel('Age constraints', fontsize = 12)
ax[1].set_yticks([])
ax[1].legend()
# sedimentation rate
age_bins = 50
rate_bins = 50
rate_df = accumulation_rate(full_trace, sample_df, ages_df, sections = section, method = 'successive', include_age_constraints = include_age_constraints_sedrate)
age_bin_edges = np.linspace(np.min(rate_df['top_age']), np.max(rate_df['base_age']), age_bins)
rate_bin_edges = np.logspace(np.log10(rate_df['rate'].min()), np.log10(rate_df['rate'].max()), rate_bins)
# get age bin centers
age_centers = (age_bin_edges[:-1] + age_bin_edges[1:]) / 2
age_vec = []
rate_vec = []
for rate, base, top in zip(rate_df['rate'], rate_df['base_age'], rate_df['top_age']):
for i in np.arange(len(age_bin_edges)-1):
center = age_centers[i]
current_bin = pd.Interval(age_bin_edges[i], age_bin_edges[i+1], closed = 'left')
current_sample_pair = pd.Interval(top, base, closed = 'right')
result = current_bin.overlaps(current_sample_pair)
if result:
age_vec.append(center)
rate_vec.append(rate)
H, xedges, yedges = np.histogram2d(age_vec, rate_vec, bins=[age_bin_edges, rate_bin_edges], density = False)
H[H == 0] = np.nan
H = H/np.nansum(H, axis=1,keepdims=1) # each column (age bin) should sum to 1
m = ax[2].pcolormesh(xedges, yedges, H.T, cmap = 'jet')
ax[2].set_yscale('log')
ax[2].set_ylabel('LOG$_{10}$ (Accumulation rate [m/Myr])', fontsize = 12)
ax[2].set_xlabel('Age (Ma)', fontsize = 12)
cb = plt.colorbar(m)
cb.set_label(label='Probability density', fontsize = 12)
# noise posteriors
if noise_type != 'none':
if noise_type == 'section':
pal = sns.color_palette("deep", n_colors = len(proxies))
if noise_type == 'groups':
palettes = ['crest', 'flare', 'Purples', 'Grays', 'Reds', 'Greens', 'YlOrBr', 'Oranges']
for i in np.arange(len(proxies)):
proxy = proxies[i]
if noise_type == 'section':
posterior_noise_dist = az.extract(full_trace.posterior)[str(section) + '_section_noise_' + proxy].values.ravel()
prior_noise_dist = az.extract(full_trace.prior)[str(section) + '_section_noise_' + proxy].values.ravel()
sns.kdeplot(posterior_noise_dist,
fill = True,
ax = ax[3],
edgecolor='none',
color = pal[i],
alpha = 0.3,
label = proxy + ' posterior',
cut = 0)
if plot_noise_prior:
sns.kdeplot(prior_noise_dist,
fill = False,
ax = ax[3],
color = pal[i],
alpha = 0.3,
label = proxy + ' prior',
cut = 0)
elif noise_type == 'groups':
noise_vars = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if f"{'_group_noise_' + str(proxy)}" in l
]
pal = sns.color_palette(palettes[i], n_colors = len(noise_vars))
for j, var in enumerate(noise_vars):
posterior_noise_dist = az.extract(full_trace.posterior)[var].values.ravel()
sns.kdeplot(posterior_noise_dist,
fill = True,
ax = ax[3],
color = pal[j],
alpha = 0.3,
label = var + ' posterior',
cut = 0
)
if plot_noise_prior:
prior_noise_dist = az.extract(full_trace.prior)[var].values.ravel()
sns.kdeplot(prior_noise_dist,
fill = False,
ax = ax[3],
color = pal[j],
alpha = 0.3,
label = var + ' prior',
cut = 0
)
if (len(proxies) > 1) | (noise_type == 'groups'):
ax[3].legend()
ax[3].set_xlabel('Noise', fontsize = 12)
ax[3].set_yticks([])
ax[3].set_ylabel(None)
# offset posteriors
if offset_type != 'none':
if offset_type == 'section':
pal = sns.color_palette("deep", n_colors = len(proxies))
if offset_type == 'groups':
palettes = ['crest', 'flare', 'Purples', 'Grays', 'Reds', 'Greens', 'YlOrBr', 'Oranges']
for i in np.arange(len(proxies)):
proxy = proxies[i]
if offset_type == 'section':
posterior_offset_dist = az.extract(full_trace.posterior)[str(section) + '_section_offset_' + proxy].values.ravel()
sns.kdeplot(posterior_offset_dist,
fill = True,
ax = ax[n_rows - 1],
edgecolor='none',
color = pal[i],
alpha = 0.3,
label = proxy + ' posterior',
cut = 0)
if plot_offset_prior:
prior_offset_dist = az.extract(full_trace.prior)[str(section) + '_section_offset_' + proxy].values.ravel()
sns.kdeplot(prior_offset_dist,
fill = False,
ax = ax[n_rows - 1],
color = pal[i],
alpha = 0.3,
label = proxy + ' prior',
cut = 0)
elif offset_type == 'groups':
offset_vars = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if f"{'_group_offset_' + str(proxy)}" in l
]
pal = sns.color_palette(palettes[i], n_colors = len(offset_vars))
for j, var in enumerate(offset_vars):
posterior_offset_dist = az.extract(full_trace.posterior)[var].values.ravel()
sns.kdeplot(posterior_offset_dist,
fill = True,
ax = ax[n_rows - 1],
edgecolor='none',
color = pal[j],
alpha = 0.3,
label = var + ' posterior',
cut = 0)
if plot_offset_prior:
prior_offset_dist = az.extract(full_trace.prior)[var].values.ravel()
sns.kdeplot(prior_offset_dist,
fill = False,
ax = ax[n_rows - 1],
color = pal[j],
alpha = 0.3,
label = var + ' prior',
cut = 0)
if (len(proxies) > 1) | (offset_type == 'groups'):
ax[n_rows - 1].legend()
ax[n_rows - 1].set_xlabel('Offset', fontsize = 12)
ax[n_rows - 1].set_yticks([])
ax[n_rows - 1].set_ylabel(None)
plt.setp(ax[0].get_xticklabels(), visible=False)
plt.setp(ax[1].get_xticklabels(), visible=False)
for axis in ax.ravel():
axis.tick_params(bottom = True, top = False, left = False, right = False, direction = 'in', labelsize = 12)
fig.suptitle('Section: ' + str(section), fontsize = 14)
return fig
[docs]
def noise_summary(full_trace, **kwargs):
"""
Plot posterior noise distributions for each section or group of samples (depending on ``noise_type`` used in :py:meth:`build_model() <stratmc.model.build_model>`) for a given proxy. If multiple proxies were included in the inference, pass a ``proxy`` argument to specify which noise terms to plot.
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import noise_summary
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
sample_df = load_object(example_sample_path)
sections = np.unique(sample_df['section'].values)
noise_summary(full_trace)
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
proxy: str, optional
Target proxy; only required if more than one proxy was included in the inference.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with noise distributions for each section or group of samples.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'proxy' in kwargs:
proxy = kwargs['proxy']
else:
proxy = proxies[0]
section_noise_vars = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if f"{'_section_noise_' + str(proxy)}" in l
]
if len(section_noise_vars) > 0:
noise_vars = section_noise_vars
else:
noise_vars = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if f"{'_group_noise_' + str(proxy)}" in l
]
if len(noise_vars) == 0:
sys.exit(f"No noise terms in the model")
N = len(noise_vars)
cols = int(np.min([4, N]))
pal = sns.color_palette("deep", n_colors = len(noise_vars))
rows = int(math.ceil(N/cols))
gs = gridspec.GridSpec(rows, cols)
fig = plt.figure(figsize = (10, 2.5*rows))
for n in range(N):
ax = fig.add_subplot(gs[n])
var = noise_vars[n]
ax.set_title(str(var), fontsize = fs)
posterior_noise_dist = full_trace.posterior[var].values.ravel()
sns.kdeplot(posterior_noise_dist,
ax = ax,
fill = True,
edgecolor='None',
color = pal[n],
alpha = 0.5,
label = proxy,
cut = 0)
ax.set_ylabel('')
ax.set_yticks([])
ax.tick_params(labelsize = fs)
fig.supxlabel('Noise (%s)' % proxy, fontsize = fs)
fig.tight_layout()
plt.subplots_adjust(wspace=0.15, hspace=0.4)
return fig
[docs]
def offset_summary(full_trace, **kwargs):
"""
Plot posterior offset distributions for each section or group of samples (depending on ``offset_type`` used in :py:meth:`build_model() <stratmc.model.build_model>`). If multiple proxies were included in the inference, pass a ``proxy`` argument to specify which offset terms to plot.
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import offset_summary
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
sample_df = load_object(example_sample_path)
offset_summary(full_trace)
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
proxy: str, optional
Target proxy; only required if more than one proxy was included in the inference.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with offset distributions for each section or group of samples.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'proxy' in kwargs:
proxy = kwargs['proxy']
else:
proxy = proxies[0]
section_offset_vars = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if f"{'_section_offset_' + str(proxy)}" in l
]
if len(section_offset_vars) > 0:
offset_vars = section_offset_vars
else:
offset_vars = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if f"{'_group_offset_' + str(proxy)}" in l
]
if len(offset_vars) == 0:
sys.exit(f"No offset terms in the model")
N = len(offset_vars)
cols = int(np.min([4, N]))
pal = sns.color_palette("deep", n_colors = len(offset_vars))
rows = int(math.ceil(N/cols))
gs = gridspec.GridSpec(rows, cols)
fig = plt.figure(figsize = (10, 2.5*rows))
for n in range(N):
ax = fig.add_subplot(gs[n])
var = offset_vars[n]
ax.set_title(var, fontsize = fs)
# shape = samples x draws
posterior_offset_dist = az.extract(full_trace.posterior)[var].values.ravel()
sns.kdeplot(posterior_offset_dist,
ax = ax,
fill = True,
edgecolor = 'None',
color = pal[n],
alpha = 0.3,
label = var,
cut = 0
)
ax.set_ylabel('')
ax.set_yticks([])
ax.tick_params(labelsize = fs)
fig.supxlabel('Offset (%s)' % proxy, fontsize = fs)
fig.tight_layout()
plt.subplots_adjust(wspace=0.15, hspace=0.4)
return fig
[docs]
def section_proxy_residuals(full_trace, sample_df, legend = True, cmap = 'Spectral', plot_mle = False, include_excluded_samples = False, **kwargs):
"""
Plot the residuals between the observed proxy values for each section and the inferred proxy signal (using the posterior section age models to map the signal back to height in section). Use to check for stratigraphic trends in the residuals, which may give insight to the processes that cause noisy sections to deviate from the inferred common signal. If multiple proxies were included in the inference, pass a ``proxy`` argument.
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import section_proxy_residuals
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
sample_df = load_object(example_sample_path)
section_proxy_residuals(full_trace, sample_df)
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
sample_df: pandas.DataFrame
:class:`pandas.DataFrame` containing proxy data for all sections.
legend: bool, optional
Generate a legend. Defaults to ``True``.
cmap: str, optional
Name of seaborn color palette to use for sections. Defaults to 'Spectral'.
plot_mle: bool, optional
Plot the maximum likelihood estimate as a line. Defaults to ``False``.
proxy: str, optional
Target proxy; only required if more than one proxy was included in the inference.
include_excluded_samples: bool, optional
Whether to plot the residuals for samples that were excluded from the inference (``Exclude?`` is ``True`` in ``sample_df``). Defaults to ``False``.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with residuals for each section.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'proxy' in kwargs:
proxy = kwargs['proxy']
else:
proxy = proxies[0]
if 'sections' in kwargs:
sections = list(kwargs['sections'])
else:
sections = np.unique(sample_df[~np.isnan(sample_df[proxy])]['section'])
cs = {}
pal = sns.color_palette(cmap, n_colors=len(sections))
N = len(sections)
cols = np.min([N, 4])
rows = int(math.ceil(N/cols))
gs = gridspec.GridSpec(rows, cols)
fig = plt.figure(figsize = (2.5 * cols, 4*rows))
proxy_pred = az.extract(full_trace.posterior_predictive)['f_pred_' + proxy].values
ages_new = full_trace.X_new.X_new.values
for n in range(N):
section = sections[n]
cs[section] = pal[n]
ax = fig.add_subplot(gs[n])
ax.set_title(section, fontsize = fs)
section_sample_df = sample_df[sample_df['section'] == section]
if include_excluded_samples:
included_idx = (~np.isnan(section_sample_df[proxy]))
else:
included_idx = (~np.isnan(section_sample_df[proxy])) & (~section_sample_df['Exclude?'])
section_ages = az.extract(full_trace.posterior)[str(section) + '_ages'].values[included_idx, :]
section_proxy_true = section_sample_df[included_idx][proxy].values
section_heights = section_sample_df[included_idx]['height'].values
section_proxy_pred = np.ones((len(section_proxy_true), proxy_pred.shape[1])) * np.nan
section_residuals = np.ones((len(section_proxy_true), proxy_pred.shape[1])) * np.nan
for i in np.arange(proxy_pred.shape[1]):
section_proxy_pred[:, i] = np.interp(section_ages[:, i], ages_new, proxy_pred[:, i])
# residual = actual - predicted
section_residuals[:, i] = section_proxy_true - section_proxy_pred[:, i]
hi = np.percentile(section_residuals, 97.5, axis = 1).flatten()
lo = np.percentile(section_residuals, 2.5, axis = 1).flatten()
ax.fill_betweenx(section_heights,
hi,
lo,
color=(.95,.95,.95),
label='95% envelope',
linestyle = 'dotted',
lw = 1.5,
edgecolor = 'k',
zorder = 0)
hi = np.percentile(section_residuals, 100-17, axis=1).flatten()
lo = np.percentile(section_residuals, 17, axis=1).flatten()
ax.fill_betweenx(section_heights,
hi,
lo,
color=(.8,.8,.8),
label='66% envelope',
linestyle = 'dashed',
lw = 1.5,
edgecolor = 'k',
zorder = 1)
hi = np.percentile(section_residuals, 100-33.5, axis=1).flatten()
lo = np.percentile(section_residuals, 33.5, axis=1).flatten()
ax.fill_betweenx(section_heights,
hi,
lo,
color=(.65,.65,.65),
label='33% envelope',
linestyle = 'solid',
lw = 1.5,
edgecolor = 'k',
zorder = 2)
if plot_mle:
# calculate mle
dy = np.linspace(np.min(section_residuals.ravel()), np.max(section_residuals.ravel()), 200)
max_like = np.zeros(section_heights.size)
for i in np.arange(section_heights.size):
height_slice = section_residuals[i,:]
max_like[i] = dy[np.argmax(gaussian_kde(height_slice, bw_method = 1)(dy))]
ax.plot(max_like,
section_heights,
color = 'k',
lw = 2,
zorder = 3,
label = 'Most likely')
ax.set_ylabel('Height (m)', fontsize = fs)
if legend and n == 0:
ax.legend(loc = 'upper right')
fig.supxlabel('Residual ' + str(proxy) + ' value (observed - predicted)', fontsize = fs)
fig.tight_layout()
return fig
[docs]
def sample_ages(full_trace, sample_df, section, plot_excluded_samples = False, plot_prior = True, cmap = 'viridis'):
"""
Plot sample age prior and posterior distributions for a given section. Each subplot contains posterior distributions for a different sample.
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import sample_ages
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
sample_df = load_object(example_sample_path)
section = '1'
sample_ages(full_trace, sample_df, section)
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of posterior and prior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
sample_df: pandas.DataFrame
:class:`pandas.DataFrame` containing proxy data for all sections.
section: str
Name of target section.
plot_excluded_samples: bool, optional
Plot age distributions for proxy observations that were excluded from the inference (``Exclude?`` is ``True`` in ``sample_df``). Defaults to ``False``.
plot_prior: bool, optional
Plot prior distributions for sample ages. Defaults to ``True``.
cmap: str, optional
Name of seaborn color palette to use for age distributions. Defaults to 'viridis'.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with prior and posterior sample age distributions.
"""
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
sample_df, _ = clean_data(sample_df, None, proxies, [section])
vals = az.extract(full_trace.posterior)[str(section)+'_ages'].values
prior_vals = az.extract(full_trace.prior)[str(section)+'_ages'].values
print(vals.shape)
print(prior_vals.shape)
if not plot_excluded_samples:
included_idx = ~sample_df['Exclude?'].values.astype(bool)
# shape = (samples x draws)
vals = vals[included_idx, :]
prior_vals = prior_vals[included_idx, :]
else:
excluded_idx = sample_df['Exclude?'].values
samples = np.mean(vals, axis = 1).size
cols = 5
N = samples
rows = int(math.ceil(N/cols))
pal = sns.color_palette(cmap,n_colors=samples)
gs = gridspec.GridSpec(rows, cols)
fig = plt.figure(figsize = (11, 1.5*rows))
excluded_count = 0
for i in range(samples):
ax = fig.add_subplot(gs[i])
if plot_excluded_samples:
if excluded_idx[i]:
linestyle = 'dashed'
edgecolor = pal[i]
post_label = 'Posterior (excluded sample)'
prior_label = 'Prior (excluded sample)'
excluded_count += 1
else:
linestyle = 'solid'
edgecolor = 'none'
post_label = 'Posterior'
prior_label = 'Prior'
else:
linestyle = 'solid'
edgecolor = 'none'
post_label = 'Posterior'
prior_label = 'Prior'
sns.kdeplot(vals[i,:].ravel(),
label = post_label, fill = True, color = pal[i], edgecolor=edgecolor, linestyle = linestyle, alpha = 0.3, ax = ax, cut = 0)
if plot_prior:
sns.kdeplot(prior_vals[i,:].ravel(),
label = prior_label, fill = False, color = pal[i], ax = ax, cut = 0)
ax.tick_params(bottom = True, top = True, left = False, right = False, direction = 'in', labelsize = 12)
ax.set_ylabel('')
ax.set_yticklabels([])
ax.invert_xaxis()
if (i == 0) | ((excluded_count == 1) and (excluded_idx[i])):
ax.legend(frameon = False, loc = 'upper left')
fig.suptitle('Section: ' + section, fontsize = 14)
fig.supxlabel('Age (Ma)', fontsize = 12)
fig.tight_layout()
return fig
[docs]
def sample_ages_per_chain(full_trace, sample_df, section, chains = None, plot_prior = False, plot_excluded_samples = False, legend = True, cmap = 'viridis'):
"""
Plot sample age posterior distributions for a given section, with separate distributions for each chain. Each subplot contains posterior distributions for a different sample. Use to check for posterior multimodality (in this example, each chain has explored a different mode of the posterior age distributions).
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import sample_ages_per_chain
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
sample_df = load_object(example_sample_path)
section = '1'
sample_ages_per_chain(full_trace, sample_df, section, chains = [0, 1, 2, 3])
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
sample_df: pandas.DataFrame
:class:`pandas.DataFrame` containing proxy data for all sections.
section: str
Name of target section.
chains: list(int) or numpy.array(int); optional
Indices of chains to include; optional (defaults to all chains).
plot_prior: bool, optional
Plot prior distributions for sample ages. Defaults to ``False``.
plot_excluded_samples: bool, optional
Plot age distributions for proxy observations that were excluded from the inference (``Exclude?`` is ``True`` in ``sample_df``). Defaults to ``False``.
legend: bool, optional
Generate a legend. Defaults to ``True``.
cmap: str, optional
Name of seaborn color palette to use for different chains. Defaults to 'viridis'.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with per-chain posterior sample age distributions.
"""
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
sample_df, _ = clean_data(sample_df, None, proxies, [section])
# chains x draws x samples
vals = full_trace.posterior[str(section)+'_ages'].values
if chains is None:
chains = np.arange(vals.shape[0])
prior_vals = full_trace.prior[str(section)+'_ages'].values
if not plot_excluded_samples:
included_idx = ~sample_df['Exclude?'].values.astype(bool)
# shape = (samples x draws)
vals = vals[:, :, included_idx]
prior_vals = prior_vals[:, :, included_idx]
excluded_idx = np.array([False] * vals.shape[2])
else:
excluded_idx = sample_df['Exclude?'].values
samples = vals.shape[2]
cols = 5
N = samples
rows = int(math.ceil(N/cols))
cs = {}
pal = sns.color_palette(cmap, n_colors=len(chains))
gs = gridspec.GridSpec(rows, cols)
fig = plt.figure(figsize = (11, 1.5*rows))
excluded_count = 0
for i in range(samples):
ax = fig.add_subplot(gs[i])
if excluded_idx[i]:
linestyle = 'dashed'
post_label = 'Posterior (excluded sample)'
prior_label = 'Prior (excluded sample)'
excluded_count += 1
lw = 1
else:
lw = 0
linestyle = 'solid'
post_label = 'Posterior'
prior_label = 'Prior'
for m in range(len(chains)):
cs[m] = pal[m]
sns.kdeplot(vals[chains[m], :, i].ravel(),
label = post_label + ', chain ' + str(chains[m]), fill = True, color = cs[m], alpha = 0.3, ax = ax, lw = lw, linestyle = linestyle, cut = 0)
# assumes only 1 prior chain
if plot_prior:
sns.kdeplot(prior_vals[:, :, i].ravel(),
label = prior_label, fill = False, color = 'k', ax = ax, cut = 0)
ax.tick_params(bottom = True, top = True, left = False, right = False, direction = 'in', labelsize = 12)
ax.set_ylabel('')
ax.set_yticklabels([])
ax.invert_xaxis()
if legend:
if (i == 0) | ((excluded_count == 1) and (excluded_idx[i])):
ax.legend(frameon = False, loc = 'upper left')
fig.suptitle('Section: ' + section, fontsize = 14)
fig.supxlabel('Age (Ma)', fontsize = 12)
fig.tight_layout()
return fig
[docs]
def age_constraints(full_trace, section, cmap = 'viridis', **kwargs):
"""
For a given section, plot prior and posterior age distributions for all depositional age constraints (and limiting age constraints that provide a minimum or maximum age for the entire section). To plot intermediate limiting ages (i.e., detrital and intrusive age constraints in the middle of a section), use :py:meth:`limiting_age_constraints() <stratmc.plotting.limiting_age_constraints>`.
.. plot::
from stratmc.data import load_trace
from stratmc.plotting import age_constraints
full_trace = load_trace('examples/example_docs_trace')
section = '1'
age_constraints(full_trace, section)
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
section: str
Name of target section.
cmap: str, optional
Name of seaborn color palette to use for age distributions. Defaults to 'viridis'.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with prior and posterior depositional age constraint distributions.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
vals = az.extract(full_trace.posterior)[str(section)+'_radiometric_age'].values
prior_vals = az.extract(full_trace.prior)[str(section)+'_radiometric_age'].values
# if only 1 age constraint (should only occur if one the min or max age is shared), reshape for plotting
# (this should actually never happen -- even if shared, the age constraints end up in the section_radiometric_age RV in the model)
if vals.ndim == 1:
vals = np.reshape(vals, (1, len(vals)))
if prior_vals.ndim == 1:
prior_vals = np.reshape(prior_vals, (1, len(prior_vals)))
samples = np.mean(vals, axis = 1).size
cols = np.min([4, samples])
N = samples
rows = int(math.ceil(N/cols))
pal = sns.color_palette(cmap, n_colors=samples)
gs = gridspec.GridSpec(rows, cols)
fig = plt.figure(figsize = (11, 2*rows))
for i in range(samples):
ax = fig.add_subplot(gs[i])
sns.kdeplot(vals[i,:].ravel(),
label = 'Posterior', fill = True, color = pal[i], edgecolor='None', alpha = 0.3, ax = ax, cut = 0)
sns.kdeplot(prior_vals[i,:].ravel(),
label = 'Prior', fill = False, color = pal[i], ax = ax, cut = 0)
ax.set_ylabel('')
ax.set_yticklabels([])
ax.invert_xaxis()
ax.tick_params(bottom = True, top = True, left = False, right = False, direction = 'in', labelsize = fs)
if i == 0:
ax.legend(frameon = False, loc = 'upper left')
fig.suptitle('Section: ' + section, fontsize = fs + 2)
fig.supxlabel('Age (Ma)', fontsize = fs)
fig.tight_layout()
return fig
[docs]
def limiting_age_constraints(full_trace, sample_df, ages_df, section, cmap = 'viridis', **kwargs):
"""
For a given section, plot prior and posterior age distributions for all intermediate limiting (i.e., detrital and intrusive ages in the middle of a section) age constraints. To plot depositional age constraints (and limiting age constraints that provide a minimum or maximum age for the entire section), use :py:meth:`age_constraints() <stratmc.plotting.age_constraints>`.
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import limiting_age_constraints
full_trace = load_trace('examples/example_docs_limiting_ages_trace')
example_sample_path = 'examples/example_sample_df_limiting_ages'
example_ages_path = 'examples/example_ages_df_limiting_ages'
sample_df = load_object(example_sample_path)
ages_df = load_object(example_ages_path)
section = '1'
limiting_age_constraints(full_trace, sample_df, ages_df, section)
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
sample_df: pandas.DataFrame
:class:`pandas.DataFrame` containing proxy data for all sections.
ages_df: pandas.DataFrame
:class:`pandas.DataFrame` containing age constraints for all sections.
section: str
Name of target section.
cmap: str, optional
Name of seaborn color palette to use for age distributions. Defaults to 'viridis'.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with prior and posterior limiting age constraint distributions.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
sample_df, ages_df = clean_data(sample_df, ages_df, proxies, [section])
detrital_df = ages_df[(ages_df['section']==section) & (ages_df['intermediate detrital?'])]
intrusive_df = ages_df[(ages_df['section']==section) & (ages_df['intermediate intrusive?'])]
detrital_named = []
if detrital_df.shape[0] > 0:
# grab names from list of model variables (will only include constriants that aren't shared)
detrital_vars = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (l[0:len(section)] == section) and (f"{'detrital_age_'}" in l)
]
detrital_named += [False] * len(detrital_vars)
# if constraints are shared, grab those names too
detrital_vars += list(detrital_df[detrital_df['shared?']]['name'].values)
detrital_named += [True] * len(list(detrital_df[detrital_df['shared?']]['name'].values))
else:
detrital_vars = []
intrusive_named = []
if intrusive_df.shape[0] > 0:
# grab names from list of model variables (will only include constriants that aren't shared)
intrusive_vars = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (l[0:len(section)] == section) and (f"{'intrusive_age_'}" in l)
]
intrusive_named += [False] * len(intrusive_vars)
# if constraints are shared, grab those names too
intrusive_vars += list(intrusive_df[intrusive_df['shared?']]['name'].values)
intrusive_named += [True] * len(list(intrusive_df[intrusive_df['shared?']]['name'].values))
else:
intrusive_vars = []
# number of age constraints to plot
combined_vars = detrital_vars + intrusive_vars
combined_named = detrital_named + intrusive_named
if len(combined_vars) == 0:
sys.exit(f"Section {section} has no intermediate limiting age constraints")
N = len(combined_vars)
cols = np.min([4, N])
rows = int(math.ceil(N/cols))
pal = sns.color_palette(cmap, n_colors=N)
gs = gridspec.GridSpec(rows, cols)
fig = plt.figure(figsize = (11, 2*rows))
for i in range(N):
var = combined_vars[i]
named = combined_named[i]
if named:
title = var
elif i < len(detrital_vars):
title = 'Detrital ' + str(i)
else:
title = 'Intrusive ' + str(i - len(detrital_vars))
ax = fig.add_subplot(gs[i])
ax.set_title(title, fontsize = fs)
# distributions for detrital and intrsive ages are always 1d (different objects created for each constraint)
vals = az.extract(full_trace.posterior)[var].values
prior_vals = az.extract(full_trace.prior)[var].values
sns.kdeplot(vals.ravel(),
label = 'Posterior', fill = True, color = pal[i], edgecolor='None', alpha = 0.3, ax = ax, cut = 0)
sns.kdeplot(prior_vals.ravel(),
label = 'Prior', fill = False, color = pal[i], ax = ax, cut = 0)
ax.set_ylabel('')
ax.set_yticklabels([])
ax.invert_xaxis()
ax.tick_params(bottom = True, top = True, left = False, right = False, direction = 'in', labelsize = fs)
if i == 0:
ax.legend(frameon = False, loc = 'upper left')
fig.suptitle('Section: ' + section, fontsize = fs + 2)
fig.supxlabel('Age (Ma)', fontsize = fs)
fig.tight_layout()
return fig
[docs]
def sadler_plot(full_trace, sample_df, ages_df, method = 'density', duration_bins = 50, rate_bins = 50, scale = 'log', include_age_constraints = True, density_cmap = 'jet', section_cmap = 'Spectral', figsize = (6, 4), **kwargs):
"""
Plot accumulation rate against duration.
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import sadler_plot
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
example_ages_path = 'examples/example_ages_df'
sample_df = load_object(example_sample_path)
ages_df = load_object(example_ages_path)
sadler_plot(full_trace, sample_df, ages_df)
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
sample_df: pandas.DataFrame
:class:`pandas.DataFrame` containing proxy data for all sections.
ages_df: pandas.DataFrame
:class:`pandas.DataFrame` containing age constraints for all sections.
method: str, optional
Plot as a 2D histogram ('density') or as a scatter plot ('scatter'). The density plot will combine data for all sections in ``sections``, while a scatter plot will assign a unique color to each section. Defaults to 'density'.
duration_bins: int, optional
Number of bin edges to use for the duration data. Defaults to 50.
rate_bins: int, optional
Number of bin edges to use for the rate data. Defaults to 50.
scale: str, optional
Scaling for x- and y-axes ('linear' or 'log'). Defaults to 'log'.
sections: list(str) numpy.array(str), optional
List of target sections. Defaults to all sections in ``sample_df``.
include_age_constraints: bool, optional
Include age constraints in sedimentation rate calculations. Defaults to ``False``.
density_cmap: str, optional
Name of matplotlib colormap to use for probability density if ``method`` is 'density'. Defaults to 'jet'.
section_cmap: str, optional
Name of seaborn color palette to use for sections if ``method`` is 'scatter'. Defaults to 'Spectral'.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with sediment accumulation rate plotted against duration.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'sections' in kwargs:
sections = list(kwargs['sections'])
else:
# any sections that have no relevant proxy observations would not have ages tracked in the trace
sections = np.unique(sample_df.dropna(subset = proxies, how = 'all')['section'])
sample_df, ages_df = clean_data(sample_df, ages_df, proxies, sections)
rate_df = accumulation_rate(full_trace, sample_df, ages_df, method = 'all', sections = sections, include_age_constraints = include_age_constraints)
dur = rate_df['duration'].values
rate = rate_df['rate'].values
fig = plt.figure(figsize = figsize)
ax = fig.gca()
if method == 'density':
if scale == 'log':
x_bins = np.logspace(np.log10(dur.min()), np.log10(dur.max()), duration_bins)
y_bins = np.logspace(np.log10(rate.min()), np.log10(rate.max()), rate_bins)
else:
x_bins = np.linspace(dur.min(), dur.max(), duration_bins)
y_bins = np.linspace(rate.min(),rate.max(), rate_bins)
H, xedges, yedges = np.histogram2d(dur, rate, bins=[x_bins, y_bins], density = False) # don't normalize during this step - calculation not correct if scale = log
H[H == 0] = np.nan
H = H/len(dur) # normalize
m = ax.pcolormesh(xedges, yedges, H.T, cmap = density_cmap)
cb = plt.colorbar(m)
cb.set_label(label='Probability density', fontsize = fs)
elif method == 'scatter':
pal = sns.color_palette(section_cmap, n_colors = len(sections))
for i in np.arange(len(sections)):
section = sections[i]
ax.scatter(rate_df[rate_df['section']==section]['duration'].values, rate_df[rate_df['section']==section]['rate'].values,
s = 5,
edgecolor = pal[i],
facecolor = 'none',
alpha = 0.5,
label = section)
ax.legend()
if scale == 'log':
ax.set_xscale('log')
ax.set_yscale('log')
ax.set_xlabel('LOG$_{10}$ (Duration [yr])', fontsize = fs)
ax.set_ylabel('LOG$_{10}$ (Accumulation rate [mm/yr])', fontsize = fs)
else:
ax.set_xlabel('Duration (yr)', fontsize = fs)
ax.set_ylabel('Accumulation rate (mm/yr)', fontsize = fs)
ax.tick_params(labelsize = fs)
fig.tight_layout()
return fig
[docs]
def accumulation_rate_stratigraphy(full_trace, sample_df, ages_df, age_bins = 50, age_bin_edges = [], rate_bins = 50, rate_bin_edges = [], rate_scale = 'log', include_age_constraints = True, cmap = 'jet', figsize = (8, 4), **kwargs):
"""
Plot the probability density of sediment accumulation rates (calculated between successive samples) through time. Probability density for each age bin sums to 1. Unless a ``sections`` argument is passed, includes accumulation rates for all sections.
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import accumulation_rate_stratigraphy
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
example_ages_path = 'examples/example_ages_df'
sample_df = load_object(example_sample_path)
ages_df = load_object(example_ages_path)
accumulation_rate_stratigraphy(full_trace, sample_df, ages_df, sections = ['1'])
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
sample_df: pandas.DataFrame
:class:`pandas.DataFrame` containing proxy data for all sections.
ages_df: pandas.DataFrame
:class:`pandas.DataFrame` containing age constraints for all sections.
age_bins: int, optional
Number of bin edges for age data (if ``age_bin_edges`` not provided). Defaults to 50.
rate_bins: int, optional
Number of bin edges for rate data (if ``rate_bin_edges`` not provided). Defaults to 50.
age_bin_edges: list(float) or numpy.array(float), optional
List or array of bin edges for the age data.
rate_bin_edges: list(float) numpy.array(float), optional
List or array of bin edges for the rate data.
rate_scale: str, optional
Scaling for rate ('linear' or 'log'). Defaults to 'log'.
include_age_constraints: bool, optional
Include age constraints in sedimentation rate calculations. Defaults to ``True``.
sections: list(str) or numpy.array(str), optional
Section(s) to include. If not provided, combines data from all sections in ``sample_df``.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with accumulation rate probability density through time.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'sections' in kwargs:
sections = list(kwargs['sections'])
else:
sections = np.unique(sample_df.dropna(subset = proxies, how = 'all')['section'])
sample_df, ages_df = clean_data(sample_df, ages_df, proxies, sections)
rate_df = accumulation_rate(full_trace, sample_df, ages_df, sections = sections, method = 'successive', include_age_constraints = include_age_constraints)
if len(age_bin_edges) == 0:
age_bin_edges = np.linspace(np.min(rate_df['top_age']), np.max(rate_df['base_age']), age_bins)
if len(rate_bin_edges) == 0:
if rate_scale == 'log':
rate_bin_edges = np.logspace(np.log10(rate_df['rate'].min()), np.log10(rate_df['rate'].max()), rate_bins)
elif rate_scale == 'linear':
rate_bin_edges = np.linspace(rate_df['rate'].min(), rate_df['rate'].max(), rate_bins)
# get age bin centers
age_centers = (age_bin_edges[:-1] + age_bin_edges[1:]) / 2
age_vec = []
rate_vec = []
for rate, base, top in zip(rate_df['rate'], rate_df['base_age'], rate_df['top_age']):
for i in np.arange(len(age_bin_edges)-1):
center = age_centers[i]
current_bin = pd.Interval(age_bin_edges[i], age_bin_edges[i+1], closed = 'left')
current_sample_pair = pd.Interval(top, base, closed = 'right')
result = current_bin.overlaps(current_sample_pair)
if result:
age_vec.append(center)
rate_vec.append(rate)
fig = plt.figure(figsize = figsize)
ax = fig.gca()
H, xedges, yedges = np.histogram2d(age_vec, rate_vec, bins=[age_bin_edges, rate_bin_edges], density = False)
H[H == 0] = np.nan
H = H/np.nansum(H, axis=1,keepdims=1) # each column (age bin) should sum to 1
m = ax.pcolormesh(xedges, yedges, H.T, cmap = cmap)
if rate_scale == 'log':
ax.set_yscale('log')
ax.set_ylabel('LOG$_{10}$ (Accumulation rate [m/Myr])', fontsize = fs)
elif rate_scale == 'linear':
ax.set_ylabel('Accumulation rate (m/Myr)', fontsize = fs)
ax.invert_xaxis()
ax.set_xlabel('Age (Ma)', fontsize = fs)
cb = plt.colorbar(m)
cb.set_label(label='Probability density', fontsize = fs)
return fig
[docs]
def section_age_range(full_trace, sample_df, ages_df, lower_age, upper_age, legend = True, section_cmap = 'Spectral', **kwargs):
"""
Plot the stratigraphic interval corresponding to a given age range (based on the posterior section age models). If ``sections`` is not provided, includes each section that overlaps the target age range.
.. plot::
from stratmc.data import load_object, load_trace
from stratmc.plotting import section_age_range
full_trace = load_trace('examples/example_docs_trace')
example_sample_path = 'examples/example_sample_df'
example_ages_path = 'examples/example_ages_df'
sample_df = load_object(example_sample_path)
ages_df = load_object(example_ages_path)
section_age_range(full_trace, sample_df, ages_df, 120, 130)
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
sample_df: pandas.DataFrame
:class:`pandas.DataFrame` containing proxy data for all sections.
ages_df: pandas.DataFrame
:class:`pandas.DataFrame` containing age constraints for all sections.
lower_age: float
Lower bound (youngest) of the target age interval.
upper_age: float
Upper bound (oldest) of the target age interval.
legend: bool, optional
Generate a legend. Defaults to ``True``.
cmap: str, optional
Name of seaborn color palette to use for sections. Defaults to 'Spectral'.
Returns
-------
fig: matplotlib.pyplot.figure
Tracer stratigraphy for each section, with the stratigraphic interval corresponding to the input age range highlighted.
"""
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'proxy' in kwargs:
proxy = kwargs['proxy']
else:
proxy = proxies[0]
if 'sections' in kwargs:
sections = list(kwargs['sections'])
else:
sections = np.unique(sample_df.dropna(subset = proxies, how = 'all')['section'])
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
sample_df, ages_df = clean_data(sample_df, ages_df, proxies, sections)
stats_df = age_range_to_height(full_trace, sample_df, ages_df,lower_age, upper_age, **kwargs)
sections = stats_df.index.tolist()
cs = {}
pal = sns.color_palette(section_cmap, n_colors=len(sections))
cols = 4
N = len(sections)
rows = int(math.ceil(N/cols))
gs = gridspec.GridSpec(rows, cols)
fig = plt.figure(figsize = (10, 4.5*rows))
for n in range(N):
section = sections[n]
cs[section] = pal[n]
ax = fig.add_subplot(gs[n])
section_df = sample_df[sample_df['section']==section]
ax.scatter(section_df[proxy], section_df['height'], color = cs[section], edgecolor = 'k', zorder = 3)
# 0 is the top height, -1 is the bottom height
ax.axhspan(stats_df.loc[section]['base_2.5'], stats_df.loc[section]['top_97.5'], zorder = 0, color = (0.95, 0.95, 0.95), label='95% envelope')
ax.axhspan(stats_df.loc[section]['base_16'], stats_df.loc[section]['top_84'], zorder = 1, color = (0.8, 0.8, 0.8), label='68% envelope')
ax.axhline(stats_df.loc[section]['base_mle'], color = 'darkgray', linestyle = 'solid', zorder = 2)
ax.axhline(stats_df.loc[section]['top_mle'], color = 'darkgray', linestyle = 'solid', zorder = 2, label = 'Most likely range')
ax.set_ylabel('Height (m)', fontsize = fs)
if proxy == 'd13c':
ax.set_xlabel(r'$\delta^{13}$C$_{\mathrm{carb}} (‰)$', fontsize = fs)
else:
ax.set_xlabel(proxy, fontsize = fs)
ax.set_title(section, fontsize = fs)
ax.tick_params(labelsize=fs)
ax.locator_params(axis='x', nbins=5)
if legend:
if n == 0:
ax.legend(loc = 'upper left')
fig.tight_layout()
plt.subplots_adjust(wspace=0.4)
return fig
[docs]
def proxy_data_gaps(full_trace, time_grid = None, yaxis = 'percentage', figsize = (6, 3.5), **kwargs):
"""
For a set of discrete time bins, shows the number of draws from the posterior where there are no proxy observations (i.e., where there are temporal `gaps` in the proxy data). This plot can be used to determine where additional observations are needed to improve the inference.
.. plot::
from stratmc.data import load_trace
from stratmc.plotting import proxy_data_gaps
full_trace = load_trace('examples/example_docs_trace')
proxy_data_gaps(full_trace, time_grid = full_trace.X_new.X_new.values[::2])
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
time_grid: np.array, optional
Time bin edges; if not provided, defaults to the ``ages`` array passed to :py:meth:`get_trace() <stratmc.inference.get_trace>`.
yaxis: str, optional
Set y-axis to percentage of posterior draws without observations ('percentage') or to the number of posterior draws without observations ('count'). Defaults to 'percentage'.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with bar plot of number of posterior draws with gaps for each time bin.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
if time_grid is None:
time_grid = full_trace.X_new.X_new.values
age_gaps, grid_centers, grid_widths, n = find_gaps(full_trace, time_grid = time_grid)
fig = plt.figure(figsize = figsize)
ax = fig.gca()
if yaxis == 'percentage':
ax.bar(grid_centers, height = 100 * age_gaps/n, width = grid_widths, color = 'lightgray', edgecolor = 'k', lw = 0.5)
ax.set_ylabel('% solutions without observations', fontsize = fs)
elif yaxis == 'count':
ax.bar(grid_centers, height = age_gaps, width = grid_widths, color = 'lightgray', edgecolor = 'k', lw = 0.5)
ax.set_ylabel('# solutions without observations', fontsize = fs)
ax.set_xlabel('Age (Ma)', fontsize = fs)
ax.tick_params(labelsize = fs)
ax.invert_xaxis()
fig.tight_layout()
return fig
[docs]
def proxy_data_density(full_trace, time_grid = None, figsize = (6, 3.5), **kwargs):
"""
Plot the mean number proxy observations in discrete time bins (averaged over all posterior draws). This plot can be used to determine where proxy observations are relatively sparse, and additional observations may improve the inference.
.. plot::
from stratmc.data import load_trace
from stratmc.plotting import proxy_data_density
full_trace = load_trace('examples/example_docs_trace')
proxy_data_density(full_trace, time_grid = full_trace.X_new.X_new.values[::2])
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
time_grid: np.array, optional
Time bin edges; if not provided, defaults to the ``ages`` array passed to :py:meth:`get_trace() <stratmc.inference.get_trace>`.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with bar plot of mean number of observations in each time bin.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
if time_grid is None:
time_grid = full_trace.X_new.X_new.values
sample_counts, grid_edges, n = count_samples(full_trace, time_grid = time_grid)
grid_widths = np.diff(grid_edges)
fig = plt.figure(figsize = figsize)
ax = fig.gca()
ax.bar(grid_edges[:-1], height = sample_counts/n, width = grid_widths, align = 'edge', color = 'lightgray', edgecolor = 'k', lw = 0.5)
ax.set_ylabel('Average # observations', fontsize = fs)
ax.set_xlabel('Age (Ma)', fontsize = fs)
ax.tick_params(labelsize = fs)
ax.invert_xaxis()
fig.tight_layout()
return fig
[docs]
def lengthscale_traceplot(full_trace, chains = None, legend = True, figsize = (5, 3.5), **kwargs):
"""
Generate trace plot (parameter value vs. step in Markov chain) for the :class:`pymc.gp.cov.ExpQuad <pymc.gp.cov.ExpQuad>` covariance kernel lengthscale. By default, includes all chains; to plot the draws for only a subset of chains, past list of chain indices to ``chains``. Use to check for posterior multimodality.
.. plot::
from stratmc.data import load_trace
from stratmc.plotting import lengthscale_traceplot
full_trace = load_trace('examples/example_docs_trace')
lengthscale_traceplot(full_trace, chains = [0, 1, 2, 3])
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
chains: list(int) or numpy.array(int); optional
Indices of chains to plot; optional (defaults to all chains).
proxy: str, optional
Target proxy; only required if more than one proxy was included in the inference.
legend: bool, optional
Generate a legend. Defaults to ``True``.
Returns
-------
fig: matplotlib.pyplot.figure
Figure with lengthscale trace plot.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'proxy' in kwargs:
proxy = kwargs['proxy']
else:
proxy = proxies[0]
fig = plt.figure(figsize = figsize)
ax = fig.gca()
if chains is None:
post_ls = full_trace.posterior['gp_ls_' +str(proxy)].values
chains = np.arange(post_ls.shape[0])
else:
post_ls = full_trace.sel(chain = chains).posterior['gp_ls_' +str(proxy)].values
sns.reset_defaults()
for i in np.arange(post_ls.shape[0]):
ax.plot(np.arange(post_ls.shape[1]),
post_ls[i, :, :],
lw = 0.5,
label = 'Chain ' + str(chains[i]))
ax.set_xlabel('Draw', fontsize = fs)
ax.set_ylabel('Lengthscale (%s)' % proxy, fontsize = fs)
if legend:
ax.legend()
ax.tick_params(labelsize = fs)
fig.tight_layout()
return fig
[docs]
def lengthscale_stability(full_trace, figsize = (5, 3.5), **kwargs):
"""
Plot the posterior standard deviation of the :class:`pymc.gp.cov.ExpQuad <pymc.gp.cov.ExpQuad>` covariance kernel lengthscale when 1 through *N* chains are considered. When the posterior has been sufficiently explored, the standard deviation will stabilize; if it has not stabilized, then additional chains should be run.
To consider chains from multiple traces associated with the same inference model, first combine the traces (saved as NetCDF files) using :py:meth:`combine_traces() <stratmc.data>` in :py:mod:`stratmc.data`.
.. plot::
from stratmc.data import load_trace
from stratmc.plotting import lengthscale_stability
full_trace = load_trace('examples/example_docs_trace')
lengthscale_stability(full_trace, fontsize = 10)
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
proxy: str, optional
Target proxy; only required if more than one proxy was included in the inference.
Returns
-------
fig: matplotlib.pyplot.figure
Figure showing the standard deviation of the covariance kernel lengthscale hyperparameter posterior for 1 through *N* chains.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'proxy' in kwargs:
proxy = kwargs['proxy']
else:
proxy = proxies[0]
gp_ls_variance = calculate_lengthscale_stability(full_trace, proxy = proxy)
fig = plt.figure(figsize = figsize)
ax = fig.gca()
ax.plot(np.arange(len(gp_ls_variance)) + 1, gp_ls_variance, color = 'k')
ax.set_xlabel('Number of chains', fontsize = fs)
ax.set_ylabel('Lengthscale standard deviation (%s)' % proxy, fontsize = fs)
ax.tick_params(labelsize = fs)
fig.tight_layout()
return fig
[docs]
def proxy_signal_stability(full_trace, figsize = (5, 3.5), **kwargs):
"""
Plot the sum (over all time slices) of the residuals between the median inferred proxy signal when all *N* chains are considered compared to when 1 to *N-1* chains are considered. When the posterior has been sufficiently explored, the residuals will stabilize and approach zero; if they have not stabilized, then additional chains should be run.
To consider chains from multiple traces associated with the same inference model, first combine the traces (saved as NetCDF files) using :py:meth:`combine_traces() <stratmc.data>` in :py:mod:`stratmc.data`.
.. plot::
from stratmc.data import load_trace
from stratmc.plotting import proxy_signal_stability
full_trace = load_trace('examples/example_docs_trace')
proxy_signal_stability(full_trace)
plt.show()
Parameters
----------
full_trace: arviz.InferenceData
An :class:`arviz.InferenceData` object containing the full set of prior and posterior samples from :py:meth:`get_trace() <stratmc.inference.get_trace>` in :py:mod:`stratmc.inference`.
proxy: str, optional
Target proxy; only required if more than one proxy was included in the inference.
Returns
-------
fig: matplotlib.pyplot.figure
Figure showing the stability of the proxy signal inference.
"""
if 'fontsize' in kwargs:
fs = kwargs['fontsize']
else:
fs = 12
# get list of proxies included in model from full_trace
variables = [
l
for l in list(full_trace["posterior"].data_vars.keys())
if (f"{'gp_ls_'}" in l) and (f"{'unshifted'}" not in l)
]
proxies = []
for var in variables:
proxies.append(var[6:])
if 'proxy' in kwargs:
proxy = kwargs['proxy']
else:
proxy = proxies[0]
median_residuals = calculate_proxy_signal_stability(full_trace, proxy = proxy)
fig = plt.figure(figsize = figsize)
ax = fig.gca()
ax.plot(np.arange(median_residuals.shape[0]) + 1, np.sum(median_residuals, axis = 1), color = 'k')
ax.set_xlabel('Number of chains', fontsize = fs)
ax.set_ylabel('Sum of median ' + str(proxy) + ' residuals', fontsize = fs)
ax.tick_params(labelsize = fs)
fig.tight_layout()
return fig
