Merge pull request #47 from minaskar/dev

infinite likelihood fix

README.md

@@ -15,7 +15,7 @@ the Preconditioned Monte Carlo (PMC) algorithm, offering significant speed impro
 traditional methods like MCMC and Nested Sampling. Ideal for large-scale scientific problems 
 with expensive likelihood evaluations, non-linear correlations, and multimodality, ``pocoMC`` 
 provides efficient and scalable posterior sampling and model evidence estimation. Widely used 
-in cosmology and astronomy, pocoMC is user-friendly, flexible, and actively maintained.
+in cosmology and astronomy, ``pocoMC`` is user-friendly, flexible, and actively maintained.
 
 ## Documentation
 

docs/source/index.rst

@@ -13,7 +13,7 @@
     traditional methods like MCMC and Nested Sampling. Ideal for large-scale scientific problems 
     with expensive likelihood evaluations, non-linear correlations, and multimodality, ``pocoMC`` 
     provides efficient and scalable posterior sampling and model evidence estimation. Widely used 
-    in cosmology and astronomy, pocoMC is user-friendly, flexible, and actively maintained.
+    in cosmology and astronomy, ``pocoMC`` is user-friendly, flexible, and actively maintained.
 
 .. admonition:: Where to start?
     :class: tip
@@ -114,6 +114,10 @@ Copyright 2022-2024 Minas Karamanis and contributors.
 Changelog
 =========
 
+**1.2.1 (14/06/24)**
+
+- Added support for log-likelihoods that return ``-np.inf`` inside the prior volume.
+
 **1.2.0 (11/06/24)**
 
 - Added ``MPIPool`` for parallelization.

pocomc/_version.py

@@ -1 +1 @@
-version = "1.2.0"
+version = "1.2.1"

pocomc/sampler.py

@@ -432,6 +432,21 @@ def run(self,
                 logl, blobs = self._log_like(x)
                 self.calls += self.n_active
 
+                # Resample prior particles with infinite likelihoods
+                inf_logl_mask = np.isinf(logl)
+                if np.any(inf_logl_mask):
+                    all_idx = np.arange(len(x))
+                    infinite_idx = all_idx[inf_logl_mask]
+                    finite_idx = all_idx[~inf_logl_mask]
+                    idx = np.random.choice(finite_idx, size=len(infinite_idx), replace=True)
+                    x[infinite_idx] = x[idx]
+                    u[infinite_idx] = u[idx]
+                    logdetj[infinite_idx] = logdetj[idx]
+                    logp[infinite_idx] = logp[idx]
+                    logl[infinite_idx] = logl[idx]
+                    if self.have_blobs:
+                        blobs[infinite_idx] = blobs[idx]
+
                 self.current_particles = dict(u=u,x=x,logl=logl,logp=logp,logdetj=logdetj,
                                     logw=-1e300 * np.ones(self.n_active), blobs=blobs, iter=self.t,
                                     calls=self.calls, steps=1, efficiency=1.0, ess=self.n_effective,