Bug fixes

environments/two_stage_train.py

@@ -1,7 +1,10 @@
 import copy
 
 from ray.rllib.env import MultiAgentEnv
-from ray.rllib.agents import ppo
+try:
+    from ray.rllib.agents import ppo
+except ImportError:
+    from ray.rllib.algorithms import ppo
 from ray.rllib.utils.framework import try_import_tf, try_import_torch
 from ray.rllib.env.apis.task_settable_env import TaskSettableEnv
 import gym

experiment_configs/cleanup-contracting.json

@@ -11,7 +11,7 @@
     "post_fcnet_hiddens": [64,64]
     ,"conv_filters":[[6,[3,3],1]]
   },
-  "wandb":true,
+  "wandb":false,
   "num_renders":5,
   "solver": true, 
   "solver_samples":100 

run_render.py

@@ -4,7 +4,10 @@
 
 """
 import copy
-from ray.rllib.agents import ppo
+try:
+    from ray.rllib.agents import ppo
+except ImportError:
+    from ray.rllib.algorithms import ppo
 from PIL import Image
 from utils.ray_config_utils import get_config_and_env
 from environments.env_utils import make_video_from_rgb_imgs

run_solver.py

@@ -7,7 +7,10 @@
 
 
 import copy
-from ray.rllib.agents import ppo
+try:
+    from ray.rllib.agents import ppo
+except ImportError:
+    from ray.rllib.algorithms import ppo
 from PIL import Image
 from utils.ray_config_utils import get_config_and_env,get_neg_config,get_solver_config
 from environments.env_utils import make_video_from_rgb_imgs

run_training.py

@@ -66,7 +66,7 @@ def ppo_learning(params_dict,config,wb_logger) :
     del config['stop_cond']
 
     analysis = ray.tune.tune.run('PPO',name=params_dict['exp_name'],stop=stop_condition,
-                 config=config,callbacks=[wb_logger],local_dir=params_dict['results_dir'],
+                 config=config,callbacks=[wb_logger],storage_path=params_dict['results_dir'],
                  num_samples=1, verbose=0 ,checkpoint_freq=10, checkpoint_at_end=True)
 
     checkpoint_paths = []

runner.py

@@ -7,7 +7,7 @@
 - `sequential_run(num_jobs, args, config_dict_list)`: Runs experiments sequentially with the specified number of jobs and configuration dictionaries.
 - `mp_run(num_jobs, total_jobs, args, config_dict_list)`: Runs experiments in parallel using multiprocessing with the specified number of jobs, total jobs, and configuration dictionaries.
 
-The script reads configuration settings from a JSON file and generates combinations of parameters for experiments. It also handles options for using GPUs and multiprocessing.
+The script reads configuration settings from a JSON filxe and generates combinations of parameters for experiments. It also handles options for using GPUs and multiprocessing.
 """
 
 import multiprocessing,time, os   
@@ -23,13 +23,13 @@ def main_run(config,args,hook_start=True,hook_end=True) :
 
 def get_args() :
     parser= argparse.ArgumentParser() 
-    parser.add_argument('--name' ,type = str , help= 'Name for campaign to run') 
+    parser.add_argument('--name' ,type = str , default='cleanup-v1') 
     parser.add_argument('--mp',action='store_true',default=False,help='Whether to use multiprocesssing')
     parser.add_argument('--workers', type = int , default=1, help = 'Only required if multiprocessing is true: Total number of workers, must be larger than the max worker requirement in the config dict') 
     parser.add_argument('--w_per_job', type = int , default=1 , help = 'Only required if multiprocessing is true: Total number of workers, must be larger than the max worker requirement in the config dict') 
     parser.add_argument('--gpu',action='store_true',default=False,help='Whether to use gpu')
-    parser.add_argument('--config_path' ,type = str , default=None , help= 'Name for campaign to run') 
-    parser.add_argument('--results_dir',type =str, default = 'results/') 
+    parser.add_argument('--config_path' ,type = str , default= 'experiment_configs/test.json') 
+    parser.add_argument('--results_dir',type =str, help='add in absolute file path', default = '~/Programming/MARL_contracts/results/') 
     parser.add_argument('--load_second_stage',type=str,default=None)
     parser.add_argument('--seeds',type=int,default=1)
 
@@ -80,6 +80,7 @@ def mp_run(num_jobs,total_jobs,args,config_dict_list) :
 
         prev_alive = alive 
         time.sleep(30) 
+
 if __name__ =='__main__' : 
     # Get arguments 
     args = get_args() 
@@ -155,7 +156,7 @@ def mp_run(num_jobs,total_jobs,args,config_dict_list) :
     config_dict_list = copy.deepcopy(new_config_list ) 
 
     if args.name is not None : 
-        args.name +=  time.strftime("_%Y-%m-%d")  
+        args.name +=  time.strftime("_%Y-%m-%d-%I:%M:%S")  
         args.results_dir += args.name
         os.mkdir(args.results_dir) 
 

utils/ray_config_utils.py

@@ -21,7 +21,7 @@
 import numpy as np
 from utils.env_creator_functions import env_creator, get_base_env_tag
 from utils.logger_utils import MetricsCallback 
-from ray.rllib.algorithms.callbacks import MultiCallbacks
+from ray.rllib.algorithms.callbacks import make_multi_callbacks
 from ray.rllib.models import ModelCatalog 
 from environments.Networks.vision_net import VisionNetwork
 
@@ -148,7 +148,7 @@ def get_config_and_env(params_dict):
                     if params_dict['shared_policy'] else (lambda agent_id, episode, worker, **kwargs: agent_id)
             },
 
-            'callbacks': MultiCallbacks([lambda : MetricsCallback()]),
+            'callbacks': make_multi_callbacks([lambda : MetricsCallback()]),
             'num_gpus': params_dict['num_gpus'],
             'stop_cond': {'timesteps_total': params_dict.get("num_timesteps")}
         }