Fix TF hanging issue in multiprocessing pools.

cli.py

@@ -1,5 +1,6 @@
 import numpy as np
-import multiprocessing
+
+from multiprocessing.pool import ThreadPool
 
 from lib.RLTrader import RLTrader
 from lib.cli.RLTraderCLI import RLTraderCLI
@@ -11,32 +12,29 @@
 args = trader_cli.get_args()
 
 
-def run_concurrent_optimize():
-    trader = RLTrader(**vars(args))
-    trader.optimize(args.trials)
-
+def run_optimize(params):
+    trader_args, logger = params
 
-def concurrent_optimize():
-    processes = []
-    for i in range(args.parallel_jobs):
-        processes.append(multiprocessing.Process(target=run_concurrent_optimize, args=()))
+    trader = RLTrader(**vars(trader_args), logger=logger)
+    trader.optimize(trader_args.trials)
 
-    print(processes)
 
-    for p in processes:
-        p.start()
+def optimize_concurrent(trader_args, logger):
+    n_processes = trader_args.parallel_jobs
 
-    for p in processes:
-        p.join()
+    opt_pool = ThreadPool(processes=n_processes)
+    opt_pool.map(run_optimize, [((trader_args, logger)) for _ in range(n_processes)])
 
 
 if __name__ == '__main__':
     logger = init_logger(__name__, show_debug=args.debug)
-    trader = RLTrader(**vars(args), logger=logger)
 
     if args.command == 'optimize':
-        concurrent_optimize()
-    elif args.command == 'train':
+        optimize_concurrent(args, logger)
+
+    trader = RLTrader(**vars(args), logger=logger)
+
+    if args.command == 'train':
         trader.train(n_epochs=args.epochs)
     elif args.command == 'test':
         trader.test(model_epoch=args.model_epoch, should_render=args.no_render)

optimize.py

@@ -1,6 +1,8 @@
-import multiprocessing
+import os
 import numpy as np
 
+from multiprocessing.pool import ThreadPool
+
 from lib.RLTrader import RLTrader
 
 np.warnings.filterwarnings('ignore')
@@ -12,18 +14,11 @@ def optimize_code(params):
 
 
 if __name__ == '__main__':
-    n_process = multiprocessing.cpu_count()
-    params = {'n_envs': n_process}
-
-    processes = []
-    for i in range(n_process):
-        processes.append(multiprocessing.Process(target=optimize_code, args=(params,)))
-
-    for p in processes:
-        p.start()
+    n_processes = 6  # os.cpu_count()
+    params = {'n_envs': n_processes}
 
-    for p in processes:
-        p.join()
+    opt_pool = ThreadPool(processes=n_processes)
+    opt_pool.map(optimize_code, [params for _ in range(n_processes)])
 
     trader = RLTrader(**params)
     trader.train(test_trained_model=True, render_trained_model=True)