Hot fix guardrails for sampling

dreamer/extraction/samplers/raycast_sampler.py

@@ -189,8 +189,30 @@ def harvest(
                 Logger.Levels.debug
             ).log()
 
+        def finalize_rays(raw_rays, target_rays):
+            lengths = np.linalg.norm(raw_rays, axis=1)
+            sorted_indices = np.argsort(lengths)
+            best_rays = raw_rays[sorted_indices][:target_rays]
+            np.random.shuffle(best_rays)
+            final_rays = best_rays
+            return final_rays
+
+        max_radius = 100
+        raw_rays = np.array([])
+
+        raddai = []
+        expansions = []
+
         while len(final_rays) < target_rays:
-            Logger(f"Sweeping lattice up to R_max = {current_R_max:.2f}...", Logger.Levels.debug).log()
+            if current_R_max >= max_radius:
+                Logger(
+                    f"[Pipeline] Could not achieve quota, found {len(final_rays)}/{target_rays}",
+                    Logger.Levels.debug
+                ).log()
+                final_rays = finalize_rays(raw_rays, target_rays)
+                break
+
+            # Logger(f"Sweeping lattice up to R_max = {current_R_max:.2f}...", Logger.Levels.debug).log()
 
             # Enforce max_per_ray=1 for the "Fair Slice"
             raw_rays = sampler.harvest(
@@ -200,15 +222,8 @@ def harvest(
             )
 
             if len(raw_rays) >= target_rays:
-                Logger(
-                    f"Quota exceeded ({len(raw_rays)}). Engaging Expanding Ball (Length Sort)...",
-                    Logger.Levels.debug
-                ).log()
-                lengths = np.linalg.norm(raw_rays, axis=1)
-                sorted_indices = np.argsort(lengths)
-                best_rays = raw_rays[sorted_indices][:target_rays]
-                np.random.shuffle(best_rays)
-                final_rays = best_rays
+                Logger(f"[Pipeline] Quota exceeded ({len(raw_rays)})!", Logger.Levels.debug).log()
+                final_rays = finalize_rays(raw_rays, target_rays)
                 break
             else:
                 if len(raw_rays) == 0:
@@ -220,15 +235,16 @@ def harvest(
 
                 # Cap the multiplier between 1.10 (minimum safety step) and 3.0 (max jump)
                 multiplier = np.clip(momentum_multiplier, 1.10, 3.0)
-                Logger(
-                    f"Discretization Gap hit: Yielded {len(raw_rays)} bounded rays. Target: {target_rays}.",
-                    Logger.Levels.debug
-                ).log()
-                Logger(
-                    f"\n   Momentum Expansion: Multiplying R_max by {multiplier:.3f}",
-                    Logger.Levels.debug
-                ).log()
+                expansions.append(multiplier)
+                raddai.append(current_R_max)
                 current_R_max *= multiplier
 
+        multipliers = [f'{radius:.2f} by {multiplier:.3f}' for multiplier, radius in zip(expansions, raddai)]
+        multipliers = ', '.join(multipliers)
+        if multipliers:
+            Logger(f"\tMomentum Expansion up to {current_R_max:.2f}: {multipliers}", Logger.Levels.debug).log()
+        else:
+            Logger(f"\tSearch radius used: {current_R_max:.2f}", Logger.Levels.debug)
+
         self._verify_uniformity(final_rays, fraction, d_flat)
         return final_rays

examples/main_example.py

@@ -46,5 +46,5 @@ def trajectory_compute_func_analysis(d):
         if_srcs=[pFq(log(2), 2, 1, -1)],
         extractor=extraction.extractor.ShardExtractorMod,
         analyzers=[analysis.AnalyzerModV1],
-        searcher=search.GeneticSearchMod
+        searcher=search.SearcherModV1
     ).run(constants=[log(2)])