New run scripts

agents.py

@@ -7,7 +7,7 @@ from steering_wheel import Controller
 class QAgent:
     gamma = 0.99
     epsilon = 1.0
-    epsilon_min = 0.01
+    epsilon_min = 0.005
     epsilon_decay = 0.9999
     online_batch_size = 64
     action_space = 1
@@ -75,6 +75,7 @@ class DQAgent(QAgent):
     def __init__(self, conf):
         super().__init__(conf)
         self.q2 = QNet(conf)
+        self.name = 'D_' + self.name
 
     def get_action(self, state):
         if np.random.rand() <= self.epsilon:

run_scripts/benchmarks.py

@@ -14,7 +14,7 @@ c.env_type = 'Lunar'
 c.net_layout = [256, 128]
 c.eps_decay = 0.9995
 c.learn_rate= 0.001
-c.run_episodes = 300
+c.run_episodes = 500
 c.save_to = 'benchmarks/'
 
 smallNet = copy.deepcopy(c)
@@ -103,10 +103,11 @@ lun.run_episodes = 500
 lun.name = 'NormalLunarDoubleNotSoMoreLearn'
 lun.net_layout = [256, 128]
 lun.conf_to_name()
-# lun.agent = QAgent(lun)
 
+
+configuration = c
 # configuration = smallNet
-configuration = smallNetSlow
+# configuration = smallNetSlow
 # configuration = smallNetDeep
 # configuration = normalNet
 # configuration = normalSlowDecay
@@ -120,10 +121,10 @@ configuration = smallNetSlow
 # configuration = deepNetSlowLearn
 # configuration = smallNetDeepSlowLearn
 # configuration = lun
-configuration = ThisHasToBeTooBig
+# configuration = ThisHasToBeTooBig
 
 print(configuration.name)
-configuration.agent = QAgent(configuration)
+configuration.agent = DQAgent(configuration)
 main.run(configuration)
 
 # configurations = [smallNet, smallNetDeep, normalNet, normalSlowDecay, normalSlowLearn, largeNet, deepNet, verryLittleNet, littleNet, verryLittleNetDeep, smallNetDeepSlowLearn, deepNetSlowLearn]

run_scripts/offline.py

+import main
+import environment_wrapper as ew
+import gym
+import copy
+import threading
+from agents import QAgent, DQAgent
+
+standard_conf = ew.Config()
+standard_conf.render = False
+standard_conf.force_cpu = True
+standard_conf.env = gym.make('LunarLander-v2')
+standard_conf.agent = None
+standard_conf.env_type = 'Lunar'
+standard_conf.name = 'StandardConf'
+standard_conf.learn = True
+standard_conf.learn_online = False
+standard_conf.learn_offline = True 
+standard_conf.net_layout= [256, 128]
+standard_conf.eps_decay = 0.6
+standard_conf.learn_rate= 0.0005
+standard_conf.learn_iterations = 4
+standard_conf.run_episodes = 100
+standard_conf.offline_epochs = 2000
+standard_conf.offline_validate_every_x_iteration = -1
+standard_conf.load_ann = False
+standard_conf.load_mem = True
+standard_conf.load_from = 'agnt'
+standard_conf.save_to = 'saved_agents/'
+
+
+test = copy.deepcopy(standard_conf)
+test.name = 'T2000'
+test.net_layout = [256, 128, 128]
+test.save_to = 'benchmarks/offline/'
+test.load_from = 'NormalDouble'
+
+
+configuration = test
+configuration.conf_to_name()
+
+configuration.agent = QAgent(configuration)
+main.run(configuration)
\ No newline at end of file

run_scripts/report.py

+import main
+import environment_wrapper as ew
+import gym
+import copy
+import threading
+from agents import QAgent, DQAgent
+
+standard_conf = ew.Config()
+standard_conf.render = False
+standard_conf.force_cpu = True
+standard_conf.env = None
+standard_conf.agent = None
+standard_conf.env_type = 'Lunar'
+standard_conf.name = 'ConfigTest'
+standard_conf.learn = True
+standard_conf.learn_online = True
+standard_conf.learn_offline = False 
+standard_conf.net_layout= [256, 128]
+standard_conf.eps_decay = 0.9996
+standard_conf.learn_rate= 0.001
+standard_conf.learn_iterations = 1
+standard_conf.run_episodes = 20
+standard_conf.offline_epochs = 1000
+standard_conf.offline_validate_every_x_iteration = 10
+standard_conf.load_ann = False
+standard_conf.load_mem = False
+standard_conf.load_from = 'agnt'
+standard_conf.save_to = 'saved_agents/'
\ No newline at end of file