Add "soft initialisation" to Sampler class

A simple change that aims to fix the issue raised in Issue #420 (closed) . Add a flag to the Sampler class init that allows for a "soft initialisation" that skips the likelihood timing and ratio verification.

Hi @michael.williams this looks good to me. Can you just let us know which exception is produced if you don't soft initialise the sampler class?

Hi @moritz.huebner, unfortunately this doesn't raise any exceptions as far as I can tell. The script just hangs but the processes remain active and don't exit unless sent a kill signal (SIGKILL).

I can put together a minimal example the replicates this behaviour if that would help.

added 1 commit

• b42d97b0 - Include _verify_parameters in soft_init

Compare with previous version

Hi @moritz.huebner, I have written a small example that replicates this behaviour (bilby_tf_example.py) but this has lead me to make a further change to the soft_init merge request.

Whilst writing this example I encountered the same behaviour as before but caused by a different function in the Sampler class, namely self._verify_parameters(). This seems to be caused by me changing the priors from an instance of bilby.gw.prior.BBHPriorDict to an instance of dict to simplify the example. It's unclear to me what in the change of priors causes the issue but given the _verify_parameters calls the self.log_likelihood() it seems that the function should be included in the soft_init.

Apologies for not catching this before the initial merge request.

Ok, I was just curious if we could deal with this using exception handling, but this is apparently not the case.

Thanks for looking into this.

approved this merge request

changed milestone to %0.5.9

added Infrastructure label

approved this merge request

merged

mentioned in commit e042803b

mentioned in issue #420 (closed)