search

sblunt / orbitize

Orbit-fitting for directly imaged objects
https://orbitize.info
Other
74 stars 43 forks source link

Running the MCMC sampler iteratively seems to increase runtime? #188

Closed vighneshnagpal closed 3 years ago

vighneshnagpal commented 3 years ago

For a given number of total steps, it appears that running the MCMC sampler iteratively drastically drives up runtime as compared to executing the run in one go. For MCMC runs of 10000 steps using the same test data, the iterative run only takes around 2000 steps in the time it takes for the non-iterative run to finish its 10000 steps.

I set up the iterations similar to the way it's documented in the Radial Velocity MCMC tutorial on the docs (code below). 

n_iter = 5 # number of iterations
for i in range(n_iter):
    # running the sampler:
    orbits = m.run_sampler(total_orbits, burn_steps=0, thin=1)
semaphoreP commented 3 years ago

We want a way to stop and resume MCMC runs for the case of long MCMC runs. Duplicate of #104