Bug

Author: RedTachyon

scripts/optimize/optuna_crowd.py

@@ -1,5 +1,6 @@
 import os
 from logging import ERROR
+from typing import Optional
 
 import cv2
 import numpy as np
@@ -37,7 +38,7 @@ class Parser(BaseParser):
     }
 
 
-def objective(trial: optuna.trial.BaseTrial, worker_id: int, path: str) -> float:
+def objective(trial: optuna.trial.BaseTrial, idx: Optional[int], worker_id: int, path: str) -> float:
     # Get some parameters
     lr = trial.suggest_loguniform("lr", 1e-5, 1e-2)
     n_episodes = 1
@@ -156,7 +157,7 @@ def objective(trial: optuna.trial.BaseTrial, worker_id: int, path: str) -> float
         entity="redtachyon",
         sync_tensorboard=True,
         config=config,
-        name=f"trial{trial.number}",
+        name=f"trial{trial.number}-{idx if idx is not None else ''}",
     )
 
     model = RelationModel(config["model"], action_space=env.action_space)
@@ -188,7 +189,7 @@ def objective(trial: optuna.trial.BaseTrial, worker_id: int, path: str) -> float
 
     # EVALUATION
     env = UnitySimpleCrowdEnv(
-        file_name=args.env,
+        file_name=path,
         virtual_display=(1600, 900),
         no_graphics=False,
         worker_id=worker_id+5,
@@ -243,21 +244,21 @@ def objective(trial: optuna.trial.BaseTrial, worker_id: int, path: str) -> float
         # Generate the dashboard
         print("Generating dashboard")
         print("Skipping dashboard")
-        # trajectory = du.read_trajectory(trajectory_path)
-        #
-        # plt.clf()
-        # du.make_dashboard(trajectory, save_path=dashboard_path)
+        trajectory = du.read_trajectory(trajectory_path)
+
+        plt.clf()
+        du.make_dashboard(trajectory, save_path=dashboard_path)
 
         # Upload to wandb
-        # print("Uploading dashboard")
-        # wandb.log(
-        #     {
-        #         "dashboard": wandb.Image(
-        #             dashboard_path,
-        #             caption=f"Dashboard {mode} {'det' if d else 'rng'} {i}",
-        #         )
-        #     }
-        # )
+        print("Uploading dashboard")
+        wandb.log(
+            {
+                "dashboard": wandb.Image(
+                    dashboard_path,
+                    caption=f"Dashboard {mode} {'det' if d else 'rng'} {i}",
+                )
+            }
+        )
 
         frame_size = renders.shape[1:3]
 
@@ -300,7 +301,7 @@ def objective(trial: optuna.trial.BaseTrial, worker_id: int, path: str) -> float
     study = optuna.load_study(storage=f"sqlite:///{args.optuna_name}.db", study_name=args.optuna_name)
 
     study.optimize(
-        lambda trial: objective(trial, args.worker_id, args.env), n_trials=args.n_trials
+        lambda trial: objective(trial, None, args.worker_id, args.env), n_trials=args.n_trials
     )
 
     print("Best params:", study.best_params)

scripts/optimize/retrain_crowd.py

@@ -44,4 +44,4 @@ class Parser(BaseParser):
         print(f"Trial {idx}")
         for i in range(args.n_trials):
             print(f"Run {i}")
-            objective(trial, args.worker_id, args.env)
+            objective(trial, i, args.worker_id, args.env)