introduce new numpy random_state, especially in store, for reproducibility

[bkmi]

For reproducibility there is the option to split the seed, we should use it.

https://numpy.org/doc/stable/reference/random/index.html?

swyft/swyft/store/store.py

Lines 94 to 131 in 07a5bff

	def add(self, N, prior, bound=None):
	"""Adds points to the store.
	Args:
	N (int): Number of samples
	prior (swyft.Prior): Prior
	bound (swyft.Bound): Bound object for prior truncation
	.. warning::
	Calling this method will alter the content of the store by adding
	additional points. Currently this cannot be reverted, so use with
	care when applying it to the DirectoryStore.
	"""
	pdf = swyft.TruncatedPrior(prior, bound)
	# Lock store while adding new points
	self.lock()
	self._update()
	# Generate new points
	z_prop = pdf.sample(N=np.random.poisson(N))
	log_lambda_target = pdf.log_prob(z_prop) + np.log(N)
	log_lambda_store = self.log_lambda(z_prop)
	log_w = np.log(np.random.rand(len(z_prop))) + log_lambda_target
	accept_new = log_w > log_lambda_store
	z_new = z_prop[accept_new]
	log_w_new = log_w[accept_new]
	# Anything new?
	if sum(accept_new) > 0:
	# Add new entries to store
	self._append_new_points(z_new, log_w_new)
	print("Store: Adding %i new samples to simulator store." % sum(accept_new))
	# Update intensity function
	self.log_lambdas.resize(len(self.log_lambdas) + 1)
	self.log_lambdas[-1] = dict(pdf=pdf.state_dict(), N=N)
	log.debug(f" total size of simulator store {len(self)}.")

[bkmi]

This would be nice for the next release, for reproducibility when dealing with multiprocessing.

[cweniger]

This would be still useful, keep it for v0.4 as optional enhancement.