change replay batch size

params/dqn_params.py

@@ -20,4 +20,4 @@ epsilon_decay = True               # Using epsilon decay
 tau_decay = True                   # Using boltzmann exploration, decay temperature
 
 replay_buffer = deque(maxlen=10000) # Size of replay buffer
-replay_batch_size = 64              # Size of replay batch
+#replay_batch_size = 64              # Size of replay batch

train_dqn.py

@@ -25,6 +25,7 @@ parser.add_argument('--lr', type=float, default=0.01, help='learning rate (defau
 parser.add_argument('--seed', type=int, default=None, help='random seed')
 parser.add_argument('--policy', type=str, choices=('epsilon_greedy', 'boltzmann'))
 parser.add_argument('--n_episode', type=int, required=True, help='number of training episodes')
+parser.add_argument('--batch_size', type=int, default=64, help='replay batch size')
 args = parser.parse_args()
 
 # Check if using cuda and define device
@@ -56,7 +57,7 @@ if args.policy == 'epsilon_greedy':
     if epsilon_decay:
         print(f'Training using epsilon decay: {epsilon_decay}')
 print(f'Size of experience replay buffer: {replay_buffer}')
-print(f'Size of experience replay batch: {replay_batch_size}')
+print(f'Size of experience replay batch: {args.batch_size}')
 print()
 
 # Initialize DQN network
@@ -123,7 +124,7 @@ for episode in range(n_episode):
             accepted_orders.append(info['Accepted orders'])
             break
 
-        loss = dqn.replay(replay_buffer, replay_batch_size, gamma)
+        loss = dqn.replay(replay_buffer, args.batch_size, gamma)
         losses.append(loss)
         state = next_state