dfm / emcee3 Goto Github PK

Allow a weighted list of moves to be provided to the Sampler.

Using emcee2 I get the following shapes after sampling:

sampler.chain: (30, 100, 3)
sampler.lnprobability: (30, 100)

Using the latest version 3.0rc2 the shapes are:

sampler.chain: (30, 100, 3)
sampler.lnprobability: (100, 30)

So the shape of chain stayed the same, but new current shape of lnprobability is not consistent anymore.

The README links to the docs for the old emcee. Also, the copyright statement in the readme seems out of date.

Hello to all,

I am running emcee 3.0rc2 with the "backends" feature and when i try to access the .h5 file i get the following error:

Traceback (most recent call last):
File "testing.py", line 23, in
samples = reader.get_chain(discard=burnin, flat=True)
File "/usr/local/lib/python3.6/dist-packages/emcee/backends/backend.py", line 72, in get_chain
return self.get_value("chain", **kwargs)
File "/usr/local/lib/python3.6/dist-packages/emcee/backends/hdf.py", line 101, in get_value
raise AttributeError("You must run the sampler with "
AttributeError: You must run the sampler with 'store == True' before accessing the results

The testing.py file contains the following:

reader = emcee.backends.HDFBackend('test_ff.h5')
fname = 'test_clust_1'
burnin = 1
samples = reader.get_chain(discard=burnin, flat=True)
bs = 40
labels = ['a','b','c','$\sigma_{e}$','k']
ax = corner.corner(samples,plot_contours = True, fill_contours = True,plot_datapoints = False, \ show_titles = True, bins = bs, levels = [0.,0.6827,0.9545,0.9973],labels=labels,title_fmt=".4f")
ax.savefig(fname+'.pdf')

The call of the run_emcee is done as follows:
if extra_force == "multithreading":
with multiprocessing.Pool() as pool:
sampler = emcee.EnsembleSampler(nwalkers, ndim, lnlikelihood, \ pool=pool,backend=backends)
start = tm.time()
sampler.run_mcmc(initial, states, progress=prog_bar)
end = tm.time()
multi_time = end - start
# print("Multiprocessing took {0:.1f} seconds".format(multi_time))

Is there anyone with the same problem? Perhaps I oversee something trivial.
Anyway, thanks for your time

This would be very useful for tuning.