Merge branch 'main' into multimodal_base

README.rst

@@ -24,12 +24,10 @@
 |github| |tests| |docs| |codecov| |pypi| |licence| |arxiv1| |arxiv2| |arxiv3| |arxiv4|
 
 
-.. |logo| image:: ./docs/assets/harm_badge_simple.svg
-    :width: 32
-    :height: 32
-    :align: center
+|logo|
 
-|logo| Harmonic
+.. |logo| image:: /docs/assets/harm_badge_simple.svg
+    :width: 90
 =================================================================================================================
 
 ``harmonic`` is an open source, well tested and documented Python implementation of the *learnt harmonic mean estimator* (`McEwen et al. 2021 <https://arxiv.org/abs/2111.12720>`_) to compute the marginal likelihood (Bayesian evidence), required for Bayesian model selection.

examples/rosenbrock.py

@@ -7,7 +7,7 @@
 
 
 def ln_prior_uniform(x, xmin=-10.0, xmax=10.0, ymin=-5.0, ymax=15.0):
-    """Compute log_e of uniform prior.
+    """Compute log_e uniform prior.
 
     Args:
 

harmonic/evidence.py

@@ -460,7 +460,7 @@ def compute_ln_inv_evidence_errors(self):
         if np.abs(ratio - 1.0) > 1e-8:
             ln_evidence_err_neg = np.log(1.0 - ratio)
         else:
-            ln_evidence_err_neg = np.NINF
+            ln_evidence_err_neg = -np.inf
 
         ln_evidence_err_pos = np.log(1.0 + ratio)
 

requirements/requirements-core.txt

@@ -1,4 +1,4 @@
-numpy
+numpy <= 1.26.4
 setuptools==68.0.0
 wheel==0.41.0
 cython>=0.29.30

tests/test_evidence.py

@@ -138,7 +138,9 @@ def test_add_chains():
 
     assert np.exp(cal_ev.ln_evidence_inv) == pytest.approx(0.159438606)
     assert np.exp(cal_ev.ln_evidence_inv_var) == pytest.approx(1.164628268e-07)
-    assert np.exp(cal_ev.ln_evidence_inv_var_var)**0.5 == pytest.approx(1.142786462e-08)
+    assert np.exp(cal_ev.ln_evidence_inv_var_var) ** 0.5 == pytest.approx(
+        1.142786462e-08
+    )
 
     nsamples1 = 300
     chains1 = ch.Chains(ndim)
@@ -155,7 +157,7 @@ def test_add_chains():
 
     assert np.exp(ev.ln_evidence_inv) == pytest.approx(0.159438606)
     assert np.exp(ev.ln_evidence_inv_var) == pytest.approx(1.164628268e-07)
-    assert np.exp(ev.ln_evidence_inv_var_var)**0.5 == pytest.approx(1.142786462e-08)
+    assert np.exp(ev.ln_evidence_inv_var_var) ** 0.5 == pytest.approx(1.142786462e-08)
 
     return
 
@@ -250,7 +252,7 @@ def test_compute_ln_inv_evidence_errors(model):
     ev.ln_evidence_inv_var = ln_ev_inv_var
 
     zeta_neg, zeta_pos = ev.compute_ln_inv_evidence_errors()
-    assert zeta_neg == np.NINF
+    assert zeta_neg == -np.inf
     assert zeta_pos == pytest.approx(np.log(2.0))
 
     # Check case where ln_ev_inv_var = ln_ev_inv
@@ -310,9 +312,7 @@ def test_compute_bayes_factors():
     ev2.chains_added = True
 
     bf12_check = ev2_inv / ev1_inv * (1.0 + ev2_inv_var / ev2_inv**2)
-    bf12_var_check = (
-        ev1_inv**2 * ev2_inv_var + ev2_inv**2 * ev1_inv_var
-    ) / ev1_inv**4
+    bf12_var_check = (ev1_inv**2 * ev2_inv_var + ev2_inv**2 * ev1_inv_var) / ev1_inv**4
 
     (bf12, bf12_std) = cbe.compute_bayes_factor(ev1, ev2)
 
@@ -335,6 +335,7 @@ def test_compute_bayes_factors():
     assert bf12 == pytest.approx(evidence)
     assert bf12_std == pytest.approx(evidence_std)
 
+
 @pytest.mark.parametrize("model", models_to_test_2)
 def test_serialization(model):
     nchains = 200