Merge branch 'private-policylearn'

Author: ct2034

multi_optim/configs.py

@@ -17,22 +17,22 @@
         "seed": 0,
         "prefix": "debug"
     },
-    "debug_no_rd": {
-        "n_nodes": 32,
-        "n_runs_pose": 8,
-        "n_runs_policy": 8,
-        "n_episodes_per_run_policy": 2,
-        "n_epochs_per_run_policy": 2,
-        "batch_size_policy": 16,
-        "stats_and_eval_every": 4,
-        "lr_pos": 1e-3,
-        "lr_policy": 1e-3,
-        "max_n_agents": 2,
-        "map_name": "x.png",
-        "seed": 0,
-        "use_reaction_diffusion": False,
-        "prefix": "debug_no_rd"
-    },
+    # "debug_no_rd": {
+    #     "n_nodes": 32,
+    #     "n_runs_pose": 8,
+    #     "n_runs_policy": 8,
+    #     "n_episodes_per_run_policy": 2,
+    #     "n_epochs_per_run_policy": 2,
+    #     "batch_size_policy": 16,
+    #     "stats_and_eval_every": 4,
+    #     "lr_pos": 1e-3,
+    #     "lr_policy": 1e-3,
+    #     "max_n_agents": 2,
+    #     "map_name": "x.png",
+    #     "seed": 0,
+    #     "use_reaction_diffusion": False,
+    #     "prefix": "debug_no_rd"
+    # },
     "tiny": {
         "n_nodes": 16,
         "n_runs_pose": 64,
@@ -93,37 +93,37 @@
         "seed": 0,
         "prefix": "large"
     },
-    "large_no_rd": {
-        "n_nodes": 128,
-        "n_runs_pose": 128,
-        "n_runs_policy": 64,
-        "n_episodes_per_run_policy": 256,
-        "n_epochs_per_run_policy": 4,
-        "batch_size_policy": 128,
-        "stats_and_eval_every": 2,
-        "lr_pos": 1e-3,  # old value when not using RD
-        "lr_policy": 1e-4,
-        "max_n_agents": 10,
-        "map_name": "x.png",
-        "seed": 0,
-        "use_reaction_diffusion": False,
-        "prefix": "large_no_rd"
-    },
-    "large_plain": {
-        "n_nodes": 128,
-        "n_runs_pose": 64,
-        "n_runs_policy": 64,
-        "n_episodes_per_run_policy": 256,
-        "n_epochs_per_run_policy": 4,
-        "batch_size_policy": 128,
-        "stats_and_eval_every": 2,
-        "lr_pos": 1e-3,
-        "lr_policy": 1e-4,
-        "max_n_agents": 10,
-        "map_name": "plain.png",
-        "seed": 0,
-        "prefix": "large_plain"
-    },
+    # "large_no_rd": {
+    #     "n_nodes": 128,
+    #     "n_runs_pose": 128,
+    #     "n_runs_policy": 64,
+    #     "n_episodes_per_run_policy": 256,
+    #     "n_epochs_per_run_policy": 4,
+    #     "batch_size_policy": 128,
+    #     "stats_and_eval_every": 2,
+    #     "lr_pos": 1e-3,  # old value when not using RD
+    #     "lr_policy": 1e-4,
+    #     "max_n_agents": 10,
+    #     "map_name": "x.png",
+    #     "seed": 0,
+    #     "use_reaction_diffusion": False,
+    #     "prefix": "large_no_rd"
+    # },
+    # "large_plain": {
+    #     "n_nodes": 128,
+    #     "n_runs_pose": 64,
+    #     "n_runs_policy": 64,
+    #     "n_episodes_per_run_policy": 256,
+    #     "n_epochs_per_run_policy": 4,
+    #     "batch_size_policy": 128,
+    #     "stats_and_eval_every": 2,
+    #     "lr_pos": 1e-3,
+    #     "lr_policy": 1e-4,
+    #     "max_n_agents": 10,
+    #     "map_name": "plain.png",
+    #     "seed": 0,
+    #     "prefix": "large_plain"
+    # },
     "mapf_benchm_random-32-32-10_debug": {
         "n_nodes": 50,
         "n_runs_pose": 4,
@@ -226,3 +226,7 @@ def augment_config_by(
 configs_all_maps = augment_config_by(configs, "map_name", [
     "c.png", "z.png"
 ])
+
+configs_more_lr_pos_s = augment_config_by(configs, "lr_pos", [
+    1e-3, 3e-4, 1e-4, 3e-5, 1e-5
+])

multi_optim/eval.py

@@ -136,8 +136,12 @@ def __init__(self, roadmap: nx.Graph, map_img: MAP_IMG,
                         this_stats["failed_cbsr"] += 1
                         continue
                     # solvable by policy
-                    res = run_a_scenario(self.first_roadmap,
-                                         agents, False, self.iterator_type)
+                    try:
+                        res = run_a_scenario(self.first_roadmap,
+                                             agents, False, self.iterator_type)
+                    except RuntimeError:
+                        this_stats["failed_policy"] += 1
+                        continue
                     if not res[IDX_SUCCESS]:
                         this_stats["failed_policy"] += 1
                         continue

multi_optim/multi_optim_run.py

@@ -28,6 +28,7 @@
 import wandb
 from cuda_util import pick_gpu_lowest_memory
 from definitions import DEFAULT_TIMEOUT_S, INVALID, MAP_IMG, PATH_W_COORDS, POS
+from multi_optim.configs import configs_more_lr_pos_s
 from planner.policylearn.edge_policy import EdgePolicyDataset, EdgePolicyModel
 from roadmaps.reaction_diffusion.example_figure import make_fig
 from roadmaps.reaction_diffusion.rd import sample_points_reaction_diffusion
@@ -297,15 +298,15 @@ def run_optimization(
     map_fname = f"roadmaps/odrm/odrm_eval/maps/{map_name}"
     if os.path.splitext(map_fname)[1] == ".png":
         map_img: MAP_IMG = read_map(map_fname)
-        map_img_inflated = inflate_map_img(map_img, radius)
+        map_img_np = np.array(map_img, dtype=np.uint8)
     elif os.path.splitext(map_fname)[1] == ".map":
-        map_np = movingai_read_mapfile(map_fname)
-        map_img = gridmap_to_map_img(map_np)
-        # lets inflate the map a bit
-        map_np_inflated = inflate_map(map_np, SUPER_RES_MULTIPLIER)
-        map_img_inflated = gridmap_to_map_img(map_np_inflated)
+        map_img_np = movingai_read_mapfile(map_fname)
+        map_img = gridmap_to_map_img(map_img_np)
     else:
         raise ValueError(f"Unknown map file extension {map_fname}")
+    map_ocv = cv2.Mat(map_img_np)
+    map_img_inflated = cv2.erode(map_ocv, np.ones(
+        (SUPER_RES_MULTIPLIER, SUPER_RES_MULTIPLIER), np.uint8))
 
     B = None
     if load_roadmap is not None:
@@ -665,27 +666,15 @@ def clear_data_folder(prefix, save_folder):
     n_processes = min(tmp.cpu_count(), 16)
     pool = tmp.Pool(processes=n_processes)
 
-    for prefix in [
-        "debug",
-        "debug_no_rd",
-        # "debug_map_name_c.png",
-        # "debug_map_name_z.png",
-        # "tiny",
-        # "tiny_map_name_c.png",
-        # "tiny_map_name_z.png",
-        # "tiny_plain",
-        # "small",
-        # "small_map_name_c.png",
-        # "small_map_name_z.png",
-        # "medium",
-        # "medium_map_name_c.png",
-        # "medium_map_name_z.png",
-        "large",
-        "large_no_rd",
-        # "large_map_name_c.png",
-        # "large_map_name_z.png",
-        # "large_plain",
-    ]:
+    configs_to_run = {
+        k: v for k, v in configs_more_lr_pos_s.items() if k.startswith("debug")
+        # or k.startswith("large")
+    }
+
+    prefixes_to_run = sorted(configs_to_run.keys())
+    print(f"Running {len(prefixes_to_run)} configs: {prefixes_to_run}")
+
+    for prefix in prefixes_to_run:
         if prefix.startswith("debug"):
             level = logging.DEBUG
         else:
@@ -703,7 +692,7 @@ def clear_data_folder(prefix, save_folder):
 
         # start the actual run
         run_optimization(
-            **configs_all_maps[prefix],
+            **configs_to_run[prefix],
             pool_in=pool)
 
     pool.close()