Unable to Perform MCMC Sample from SyntheticData.py

[ngierty]

I adapted the code in the SyntheticData.py to run in Spyder. However, the radvel.mcmc step fails with "IndexError: list index out of range". I'm running Python 3.9.7 with Spyder 5.4.2 on Mac OSX Ventura (13.0).

Here's the (slightly) modified code to reproduce the issue:

`''' Synthetic data tutorial for RadVel '''

import radvel import radvel.likelihood import numpy as np import matplotlib.pyplot as plt import copy from scipy import optimize

Initialize Keplerian Model Parameters

params = radvel.Parameters(1,basis='per tc secosw sesinw k') params['k1'] = radvel.Parameter(value=1) params['per1'] = radvel.Parameter(value=1) params['secosw1'] = radvel.Parameter(value=0.1) params['sesinw1'] = radvel.Parameter(value=+0.2) params['tc1'] = radvel.Parameter(value=0.) rv_mod = radvel.RVModel(params)

Generate Synthetic Data

t = np.random.random_sample(40) t = t * 4 - 2 t = np.sort(t) ti = np.linspace(-2,2,400) errvel = 0.3 jitter = 0.3 syn_rv = rv_mod(t) + np.random.normal(scale = np.sqrt(errvel2 + jitter2), size = t.size) plt.errorbar(t,syn_rv,yerr=errvel,fmt='.',label='Synthetic Data') plt.plot(ti,rv_mod(ti),label='Underlying Model') plt.xlabel('Time') plt.ylabel('RV') plt.legend() plt.show()

Generate the likelihood

like_syn = radvel.likelihood.RVLikelihood(rv_mod,t,syn_rv,np.zeros(t.size)+errvel) like_syn.params['gamma'] = radvel.Parameter(value=0) like_syn.params['jit'] = radvel.Parameter(value=errvel)

truths = copy.deepcopy(like_syn.params) # Store away model parameters for later reference like_syn.params.update(dict(k1=radvel.Parameter(value=3), secosw1=radvel.Parameter(value=0.1), sesinw1=radvel.Parameter(value=0.1), tc1=radvel.Parameter(value=0.1))) # perturb the starting guess

like_syn.params['jit'].vary = False # Don't vary jitter

Perform a maximum likelihood fit

res = optimize.minimize(like_syn.neglogprob_array, like_syn.get_vary_params(), method='Nelder-Mead' )

res = optimize.minimize(like_syn.neglogprob_array, like_syn.get_vary_params(), method='L-BFGS-B' )

print(res) print(like_syn) plt.errorbar(like_syn.x, like_syn.model(t)+like_syn.residuals(), yerr=like_syn.yerr, fmt='o') plt.plot(ti, like_syn.model(ti)) plt.xlabel('Time') plt.ylabel('RV') plt.show()

Instantiate a posterior object

post = radvel.posterior.Posterior(like_syn) post.params['per1'] = radvel.Parameter(value=1) post.params['k1'] = radvel.Parameter(value=1) post.params['jit'].vary = True post.priors += [radvel.prior.EccentricityPrior( 1 )] post.priors += [radvel.prior.Gaussian( 'jit', errvel, 0.1)] post.priors += [radvel.prior.Gaussian( 'per1', 1, 0.1)] post.priors += [radvel.prior.Gaussian( 'tc1', 0, 0.1)]

post.priors += [radvel.prior.SecondaryEclipsePrior(1, 0.5, 0.01)]

print(post)

Perform a maximum likelihood fit on posterior

res = optimize.minimize( post.neglogprob_array, post.get_vary_params(), method='Powell', options=dict(maxiter=100000,maxfev=100000,xtol=1e-8) )

plt.errorbar(post.likelihood.x, post.likelihood.model(t)+like_syn.residuals(), yerr=post.likelihood.yerr, fmt='o') plt.plot(ti, post.likelihood.model(ti)) plt.xlabel('Time') plt.ylabel('RV') plt.show()

Use MCMC to sample the posterior

df = radvel.mcmc(post, nrun=200, savename='rawchains.h5')`

[bjfultn]

Could you post the full traceback?

[ngierty]

Here's a screenshot of the traceback I get.

[bjfultn]

This one is really stumping me. Can you run the other tutorials ok?

[ngierty]

Sorry for the delay. I have tried the Custom Model Tutorial but get the same issue on Spyder.

I'm able to somewhat get around the issue by following the example in the published paper and performing the fit in the command line. However, it is unclear to me how I would implement a custom likelihood if I can't perform the MCMC fit inside of Spyder.

[ngierty]

I think this may have been an install issue on my end somewhere. In between doing a new installation of anaconda and reinstalling RadVel the issue cleared up. If I come across it again, I'll reopen the issue with hopefully some more insight.