Hi everybody!
In the case of writing new distribution in pymc3, I’m trying to write a defined distribution in the program to compare its result with pymc3 itself (in this way, I can extend the strategy to undefined distributions). For this approach, my choice was exponential distribution and my code is:
Blockquote
import theano.tensor as tt
def expo(lam):
def logp(x):
return tt.log(lam) -lam*x
return logp
with pm.Model():
s = pm.Uniform(‘s’,0,10)
obs = pm.DensityDist(‘obs’,expo(s),observed = 3)
trace = pm.sample(20000)
with pm.Model():
s1 = pm.Uniform(‘s1’,0,10)
obs1 = pm.Exponential(‘obs1’,lam = s1,observed = 3)
trace1 = pm.sample(20000)
sns.distplot(trace[‘s’])
sns.distplot(trace1[‘s1’])
plt.legend([‘khodam’,‘py’])
plt.show()
when I run this code, suddenly my kernel dies!! with this message on the display: ‘The kernel appears to have died. It will restart automatically’. Would you please help me to solve this?!
Thank you
1 Like
Are you on Windows? There are common issues with parallel sampling on Windows that could be behind the kernel dying. You can try setting cores = 1 in pm.sample and see it solves it.
2 Likes
yes, my OS is windows and (again) yess it worked!.. thank you
Could you tell me what is the role of ‘cores’ in the code?? why should I give 1 to this value?
1 Like
Cores means how many cpu cores will be used for sampling. By default this is 4, meaning PyMC3 will sample 4 chains in parallel. Unfortunately, this often fails on Windows. If you set it cores to 1, PyMC3 will still sample 4 chains, but one at a time.
2 Likes
Hi, I’m facing this same error with a code I was previously able to run before.
Setting up the cores=1 didn’t solve the problem
Im’ using Ubuntu in an virtual box in WIndows
Are you running in a jupyter notebook? If so, you might consider running it as an executable script from the command line as there are often more informative error messages that way. Otherwise, you can try to provide a short example of code that reproduces the error.