Fix numerical instability in tutorial likelihood

Expected counts of zero result in NaN log-likelihood gradient.
bat · Oct 20, 2023 · 3e10e78 · 3e10e78
1 parent c6d2c65
commit 3e10e78
Showing 1 changed file with 2 additions and 1 deletion.
diff --git a/docs/src/tutorial_lit.jl b/docs/src/tutorial_lit.jl
@@ -163,7 +163,8 @@ likelihood = let h = hist, f = fit_function
         function bin_log_likelihood(i)
             ## Simple mid-point rule integration of fit function `f` over bin:
             expected_counts = bin_widths[i] * f(params, bin_centers[i])
-            logpdf(Poisson(expected_counts), observed_counts[i])
+            # Avoid zero expected counts for numerical stability:
+            logpdf(Poisson(expected_counts + eps(expected_counts)), observed_counts[i])
         end
 
         ## Sum log-likelihood over bins: