RuntimeWarning: invalid value encountered in double_scalars

General information:

• emcee version: 2.2.1
• platform: Ubuntu LTS 18.04
• installation method (pip/conda/source/other?): pip

Problem description:

Expected behavior:

No error

Actual behavior:

I get a RuntimeWarning: invalid value encountered in double_scalars in the red_blue.py script at line 97: lnpdiff = f + nlp - state.log_prob[j]. The same error appears from the old ensemble.py script at the line: lnpdiff = (self.dim - 1.) * np.log(zz) + newlnprob - lnprob0 (Issue 269, now closed)

What have you tried so far?:

I have tried both versions of the code, and I can’t figure out why I get that error. For an Ensemble with a lower amount of steps I do not get that error, only when the chain gets longer.

Minimal example:

import emcee



# sample code goes here...

[dfm]

This generally means that you are initializing in some part of parameter space where the log probability is -inf or nan. Double check that your initialized values are all ok. If they are, then there’s a chance that this is caused by your log probability function returning nan or inf, but that should be caught in the sampler. One way or another you should always try to catch issues like that in your log probability function and return -np.inf. Give that a shot.

[ayushp88]

I’m not sure what you mean by “control”, but this does mean that there’s some numerical issue happening in your log probability function and it normally has to do with initialization. I’d recommend trying to make a simple test case that reproduces the problem. That should make it clear what’s going on.

Sir I am also facing the same issue, can you please help me.