Merge pull request #131 from EMI-Group/fix-cec2022

fix: avoid nan

src/evox/problems/numerical/cec2022_so.py

@@ -127,7 +127,8 @@ def compose_operat(self, x, fs, Os_mat, bias, sigma, lamb):
         term1 = 1 / jnp.sqrt(diff_square_sum)
         term2 = jnp.exp(-0.5 * diff_square_sum / (sigma**2 * D))
         W = term1 * term2
-        W_norm = W / jnp.sum(W)
+        term1_nan = jnp.isnan(term1)
+        W_norm = lax.select(jnp.any(term1_nan), term1_nan / jnp.count_nonzero(term1_nan), W / jnp.sum(W))
 
         f = jnp.sum(W_norm * (lamb * fs + bias))