Skip to content

Add average along-chain residual to mcmc diagnostics plot #157

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
btobers merged 2 commits into from
Oct 30, 2025
Merged

Add average along-chain residual to mcmc diagnostics plot #157

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
26f1dc6
Add average along-chain residual to mcmc diagnostics plot for non-gla…
btobers Oct 30, 2025
4da600d
Vertically align ylabels
btobers Oct 30, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 4 additions & 11 deletions pygem/bin/run/run_calibration.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2219,21 +2219,12 @@ def rho_constraints(**kwargs):
f'Chain {n_chain}: failed to produce an unstuck result after {attempts_per_chain} initial guesses.'
)

# concatenate mass balance
m_chain = torch.cat(
(m_chain, torch.tensor(pred_chain['glacierwide_mb_mwea']).reshape(-1, 1)), dim=1
)
m_primes = torch.cat(
(m_primes, torch.tensor(pred_primes['glacierwide_mb_mwea']).reshape(-1, 1)),
dim=1,
)

if debug:
print(
'mb_mwea_mean:',
np.round(torch.mean(m_chain[:, -1]).item(), 3),
np.round(torch.mean(torch.stack(pred_chain['glacierwide_mb_mwea'])).item(), 3),
'mb_mwea_std:',
np.round(torch.std(m_chain[:, -1]).item(), 3),
np.round(torch.std(torch.stack(pred_chain['glacierwide_mb_mwea'])).item(), 3),
'\nmb_obs_mean:',
np.round(mb_obs_mwea, 3),
'mb_obs_std:',
Expand All @@ -2257,6 +2248,8 @@ def rho_constraints(**kwargs):
graphics.plot_mcmc_chain(
m_primes,
m_chain,
pred_primes,
pred_chain,
obs,
ar,
glacier_str,
Expand Down
72 changes: 55 additions & 17 deletions pygem/plot/graphics.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,6 +13,7 @@

import matplotlib.pyplot as plt
import numpy as np
import torch
from scipy.stats import binned_statistic

from pygem.utils.stats import effective_n
Expand Down Expand Up @@ -113,10 +114,15 @@ def plot_modeloutput_section(model=None, ax=None, title='', **kwargs):
ax.set_title(title, loc='left')


def plot_mcmc_chain(m_primes, m_chain, obs, ar, title, ms=1, fontsize=8, show=False, fpath=None):
def plot_mcmc_chain(
m_primes, m_chain, pred_primes, pred_chain, obs, ar, title, ms=1, fontsize=8, show=False, fpath=None
):
# Plot the trace of the parameters
n = m_primes.shape[1]
fig, axes = plt.subplots(n + 1, 1, figsize=(6, n * 1.5), sharex=True)
nparams = m_primes.shape[1]
npreds = len(pred_chain.keys())
N = nparams + npreds + 1
fig, axes = plt.subplots(N, 1, figsize=(6, N * 1), sharex=True)
# convert torch objects to numpy
m_chain = m_chain.detach().numpy()
m_primes = m_primes.detach().numpy()

Expand All @@ -125,6 +131,7 @@ def plot_mcmc_chain(m_primes, m_chain, obs, ar, title, ms=1, fontsize=8, show=Fa
# instantiate list to hold legend objs
legs = []

# axes[0] will always be tbias
axes[0].plot(
[],
[],
Expand All @@ -139,6 +146,7 @@ def plot_mcmc_chain(m_primes, m_chain, obs, ar, title, ms=1, fontsize=8, show=Fa
# axes[0].add_artist(leg)
axes[0].set_ylabel(r'$T_{bias}$', fontsize=fontsize)

# axes[1] will always be kp
axes[1].plot(m_primes[:, 1], '.', ms=ms, c='tab:blue')
axes[1].plot(m_chain[:, 1], '.', ms=ms, c='tab:orange')
axes[1].plot(
Expand All @@ -150,6 +158,7 @@ def plot_mcmc_chain(m_primes, m_chain, obs, ar, title, ms=1, fontsize=8, show=Fa
legs.append(l1)
axes[1].set_ylabel(r'$K_p$', fontsize=fontsize)

# axes[2] will always be ddfsnow
axes[2].plot(m_primes[:, 2], '.', ms=ms, c='tab:blue')
axes[2].plot(m_chain[:, 2], '.', ms=ms, c='tab:orange')
axes[2].plot(
Expand All @@ -161,7 +170,8 @@ def plot_mcmc_chain(m_primes, m_chain, obs, ar, title, ms=1, fontsize=8, show=Fa
legs.append(l2)
axes[2].set_ylabel(r'$fsnow$', fontsize=fontsize)

if n > 4:
if nparams > 3:
# axes[3] will be rho_ablation if more than 3 model params
m_chain[:, 3] = m_chain[:, 3]
m_primes[:, 3] = m_primes[:, 3]
axes[3].plot(m_primes[:, 3], '.', ms=ms, c='tab:blue')
Expand All @@ -175,6 +185,7 @@ def plot_mcmc_chain(m_primes, m_chain, obs, ar, title, ms=1, fontsize=8, show=Fa
legs.append(l3)
axes[3].set_ylabel(r'$\rho_{abl}$', fontsize=fontsize)

# axes[4] will be rho_accumulation if more than 3 model params
m_chain[:, 4] = m_chain[:, 4]
m_primes[:, 4] = m_primes[:, 4]
axes[4].plot(m_primes[:, 4], '.', ms=ms, c='tab:blue')
Expand All @@ -188,33 +199,60 @@ def plot_mcmc_chain(m_primes, m_chain, obs, ar, title, ms=1, fontsize=8, show=Fa
legs.append(l4)
axes[4].set_ylabel(r'$\rho_{acc}$', fontsize=fontsize)

if 'glacierwide_mb_mwea' in obs.keys():
# plot predictions
if 'glacierwide_mb_mwea' in pred_primes.keys():
mb_obs = obs['glacierwide_mb_mwea']
axes[-2].fill_between(
axes[nparams].fill_between(
np.arange(len(ar)),
mb_obs[0] - (2 * mb_obs[1]),
mb_obs[0] + (2 * mb_obs[1]),
color='grey',
alpha=0.3,
)
axes[-2].fill_between(
axes[nparams].fill_between(
np.arange(len(ar)),
mb_obs[0] - mb_obs[1],
mb_obs[0] + mb_obs[1],
color='grey',
alpha=0.3,
)
axes[-2].plot(m_primes[:, -1], '.', ms=ms, c='tab:blue')
axes[-2].plot(m_chain[:, -1], '.', ms=ms, c='tab:orange')
axes[-2].plot(

mb_primes = torch.stack(pred_primes['glacierwide_mb_mwea']).numpy()
mb_chain = torch.stack(pred_chain['glacierwide_mb_mwea']).numpy()
axes[nparams].plot(mb_primes, '.', ms=ms, c='tab:blue')
axes[nparams].plot(mb_chain, '.', ms=ms, c='tab:orange')
axes[nparams].plot(
[],
[],
label=f'median={np.median(m_chain[:, -1]):.3f}\niqr={np.subtract(*np.percentile(m_chain[:, -1], [75, 25])):.3f}',
label=f'median={np.median(mb_chain):.3f}\niqr={np.subtract(*np.percentile(mb_chain, [75, 25])):.3f}',
)
ln2 = axes[-2].legend(loc='upper right', handlelength=0, borderaxespad=0, fontsize=fontsize)
ln2 = axes[nparams].legend(loc='upper right', handlelength=0, borderaxespad=0, fontsize=fontsize)
legs.append(ln2)
axes[-2].set_ylabel(r'$\dot{{b}}$', fontsize=fontsize)
axes[nparams].set_ylabel(r'$\dot{{b}}$', fontsize=fontsize)
nparams += 1

# plot MAE for all other prediction keys
for key in pred_primes.keys():
if key == 'glacierwide_mb_mwea':
continue
pred_primes = torch.stack(pred_primes[key]).numpy()
pred_chain = torch.stack(pred_chain[key]).numpy()
obs_vals = np.array(obs[key][0])

mae_primes = np.mean(pred_primes - obs_vals, axis=(1, 2))
mae_chain = np.mean(pred_chain - obs_vals, axis=(1, 2))

axes[nparams].plot(mae_primes, '.', ms=ms, c='tab:blue')
axes[nparams].plot(mae_chain, '.', ms=ms, c='tab:orange')

if key == 'elev_change_1d':
axes[nparams].set_ylabel(r'$\overline{\hat{dh} - dh}$', fontsize=fontsize)
else:
axes[nparams].set_ylabel(r'$\overline{\mathrm{pred} - \mathrm{obs}}$', fontsize=fontsize)
legs.append(None)
nparams += 1

# axes[-1] will always be acceptance rate
axes[-1].plot(ar, 'tab:orange', lw=1)
axes[-1].plot(
np.convolve(ar, np.ones(100) / 100, mode='valid'),
Expand All @@ -230,7 +268,8 @@ def plot_mcmc_chain(m_primes, m_chain, obs, ar, title, ms=1, fontsize=8, show=Fa
ax.xaxis.set_ticks_position('both')
ax.yaxis.set_ticks_position('both')
ax.tick_params(axis='both', direction='inout')
if i == n:
ax.yaxis.set_label_coords(-0.1, 0.5)
if i > m_primes.shape[1] - 1:
continue
ax.plot([], [], label=f'n_eff={neff[i]}')
hands, ls = ax.get_legend_handles_labels()
Expand All @@ -252,9 +291,8 @@ def plot_mcmc_chain(m_primes, m_chain, obs, ar, title, ms=1, fontsize=8, show=Fa
handlelength=0,
fontsize=fontsize,
)

for i, ax in enumerate(axes):
ax.add_artist(legs[i])
if legs[i] is not None:
ax.add_artist(legs[i])

axes[0].set_xlim([0, m_chain.shape[0]])
axes[0].set_title(title, fontsize=fontsize)
Expand Down
Loading