gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-17 11:10:20 +01:00
Code Issues Projects Releases Wiki Activity

bug fix - updating Doom_Health_DFP and Breakout_DQN presets

Browse Source
This commit is contained in:
Itai Caspi
2017-10-31 10:54:14 +02:00
parent 913ab75e8a
commit e38611b9eb
1 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
presets.py
View File

@@ -202,6 +202,7 @@ class Doom_Health_DFP(Preset):
self.learning_rate = 0.00025 self.learning_rate = 0.00025
self.num_heatup_steps = 1000 self.num_heatup_steps = 1000
self.exploration.epsilon_decay_steps = 10000 self.exploration.epsilon_decay_steps = 10000
self.agent.use_accumulated_reward_as_measurement = True
class Doom_Deadly_Corridor_Bootstrapped_DQN(Preset): class Doom_Deadly_Corridor_Bootstrapped_DQN(Preset):
@@ -213,6 +214,7 @@ class Doom_Deadly_Corridor_Bootstrapped_DQN(Preset):
self.agent.num_steps_between_copying_online_weights_to_target = 1000 self.agent.num_steps_between_copying_online_weights_to_target = 1000
self.num_heatup_steps = 1000 self.num_heatup_steps = 1000
class CartPole_Bootstrapped_DQN(Preset): class CartPole_Bootstrapped_DQN(Preset):
def __init__(self): def __init__(self):
Preset.__init__(self, BootstrappedDQN, GymVectorObservation, BootstrappedDQNExploration) Preset.__init__(self, BootstrappedDQN, GymVectorObservation, BootstrappedDQNExploration)
@@ -371,8 +373,8 @@ class Breakout_DQN(Preset):
self.exploration.evaluation_policy = 'EGreedy' self.exploration.evaluation_policy = 'EGreedy'
self.exploration.evaluation_epsilon = 0.05 self.exploration.evaluation_epsilon = 0.05
self.num_heatup_steps = 50000 self.num_heatup_steps = 50000
self.evaluation_episodes = 25 self.evaluation_episodes = 1
self.evaluate_every_x_episodes = 1000 self.evaluate_every_x_episodes = 100
class Breakout_DistributionalDQN(Preset): class Breakout_DistributionalDQN(Preset):