Fix GA crashing

dreamer/configs/search.py

@@ -50,6 +50,9 @@ class SearchConfig(Configurable):
     GA_MAX_RETRIES: int = 3         # Retry rounds for invalid/failed trajectory evaluations.
     GA_REFINE_PROB: float = 0.5     # Probability of entering refine mutation mode.
     GA_REFINE_COORD_PROB: float = 0.5   # Per-coordinate refine perturbation probability.
+    GA_MAX_NO_IMPROVEMENT_COUNT_RETRY: int = 5  # Max retries when no improvement is observed before giving up.
+
+    MAX_TRAJECTORY_COORD: int = 50  # Max coordinate value for a trajectory.
 
 
 search_config: SearchConfig = SearchConfig()

dreamer/extraction/samplers/raycast_sampler.py

@@ -3,8 +3,11 @@
 from dreamer.extraction.samplers.conditioner import HyperSpaceConditioner
 from dreamer.extraction.samplers.raycaster import RayCastingSamplingMethod
 from dreamer.utils.logger import Logger
+from dreamer.configs.search import search_config
 from .sampler import Sampler
 from typing import Callable, cast
+import math
+
 
 
 class RaycastPipelineSampler(Sampler):
@@ -197,7 +200,7 @@ def finalize_rays(raw_rays, target_rays):
             final_rays = best_rays
             return final_rays
 
-        max_radius = 100
+        max_radius = math.sqrt(pow(search_config.MAX_TRAJECTORY_COORD, 2) * d_flat) + 1
         raw_rays = np.array([])
 
         raddai = []
@@ -212,9 +215,6 @@ def finalize_rays(raw_rays, target_rays):
                 ).log()
                 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(
                 target_rays=guide_rays_to_shoot,
                 R_max=current_R_max,

dreamer/search/methods/genetic.py

@@ -282,7 +282,10 @@ def _sample_valid_trajectories(self, *, count: int, template_pos: Position) -> L
                         break
 
         if len(sampled) < count:
-            raise ValueError("Genetic search could not sample enough valid trajectories satisfying A v <= 0")
+            Logger(
+                f"Genetic search could not sample enough valid trajectories. Sampled {len(sampled)}/{count}",
+                Logger.Levels.warning
+            ).log()
         return sampled[:count]
 
     def _get_valid_repair_trajectory(self, template_pos: Position) -> Position:
@@ -293,7 +296,8 @@ def _get_valid_repair_trajectory(self, template_pos: Position) -> Position:
                 count=self._buffer_chunk_size,
                 template_pos=template_pos
             )
-            random.shuffle(self._valid_trajectory_buffer)
+            if self._valid_trajectory_buffer:
+                random.shuffle(self._valid_trajectory_buffer)
         return self._valid_trajectory_buffer.pop()
 
     def _repair_trajectory(self, trajectory: Position, template_pos: Position) -> Position:
@@ -406,11 +410,28 @@ def _evaluate_population(
                 population[i]["sd"] = sd
 
         unresolved = invalid_indices
+        unchanged_count = 0
+        last_found_amount = -1
+
         for _ in range(self.max_retries):
             if not unresolved:
                 break
 
+            if unchanged_count >= search_config.GA_MAX_NO_IMPROVEMENT_COUNT_RETRY:
+                Logger(
+                    "Genetic algorithm solving unresolved trajectories - giving up resampling.", Logger.Levels.debug
+                ).log()
+                break
+
             retry_trajectories = self._sample_valid_trajectories(count=len(unresolved), template_pos=template_pos)
+            if len(retry_trajectories) == 0:
+                Logger("No valid trajectories could be sampled")
+
+            if last_found_amount == -1:
+                last_found_amount = len(retry_trajectories)
+            elif last_found_amount == len(retry_trajectories):
+                unchanged_count += 1
+
             retry_pairs = [(traj, start) for traj in retry_trajectories]
             self._compute_missing_search_data(retry_pairs)
 
@@ -431,7 +452,6 @@ def _evaluate_population(
                     next_unresolved.append(i)
 
             unresolved = next_unresolved
-
         return population
 
     def search(
@@ -450,6 +470,10 @@ def search(
         template = self._resolve_template(template_trajectory)
 
         initial_trajectories = self._sample_valid_trajectories(count=self.pop_size, template_pos=template)
+        if len(initial_trajectories) == 0:
+            Logger("No valid trajectories could be sampled. Continue...", Logger.Levels.warning).log()
+            return DataManager(self.use_LIReC)
+
         population: List[Dict[str, Any]] = [
             {"trajectory": traj, "delta": None, "sd": None} for traj in initial_trajectories
         ]

tests/test_search_genetic.py

@@ -375,24 +375,6 @@ def _always_invalid_mutate(_pos, **_kwargs):
     assert all(space.is_valid_trajectory(sd.sv.trajectory) for sd in result.values())
 
 
-def test_genetic_search_raises_when_constraints_have_no_valid_trajectories(monkeypatch):
-    """Ensure search fails loudly when constraints admit no valid trajectories.
-    Assumption: sampler yields only invalid trajectories for constrained space.
-    Failure mode caught: infinite retries or misleading exception messages.
-    """
-    space = ImpossibleConstrainedSpace()
-    x, y = space.cmf.symbols
-    _patch_static_sampler(monkeypatch, [Position({x: 1, y: 1}), Position({x: 2, y: 2})])
-    _configure_ga(monkeypatch, generations=1, pop_size=2, max_retries=1, parallel_search=False)
-    method = GeneticSearchMethod(
-        cast(Searchable, cast(object, space)),
-        constant=None,
-    )
-
-    with pytest.raises(ValueError, match="could not sample enough valid trajectories"):
-        method.search(template_trajectory=Position({x: 1, y: 1}))
-
-
 def test_evaluate_population_resamples_invalid_trajectories_in_batch(monkeypatch):
     """Verify invalid individuals are resampled in one batch per retry round.
     Assumption: initial cache lookup returns no SearchData for all individuals.