From e38611b9eb80d86b9b8e26b7feded9ba5aaba4a8 Mon Sep 17 00:00:00 2001 From: Itai Caspi Date: Tue, 31 Oct 2017 10:54:14 +0200 Subject: [PATCH] bug fix - updating Doom_Health_DFP and Breakout_DQN presets --- presets.py | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/presets.py b/presets.py index 96e8de7..51bba41 100644 --- a/presets.py +++ b/presets.py @@ -202,6 +202,7 @@ class Doom_Health_DFP(Preset): self.learning_rate = 0.00025 self.num_heatup_steps = 1000 self.exploration.epsilon_decay_steps = 10000 + self.agent.use_accumulated_reward_as_measurement = True 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.num_heatup_steps = 1000 + class CartPole_Bootstrapped_DQN(Preset): def __init__(self): Preset.__init__(self, BootstrappedDQN, GymVectorObservation, BootstrappedDQNExploration) @@ -371,8 +373,8 @@ class Breakout_DQN(Preset): self.exploration.evaluation_policy = 'EGreedy' self.exploration.evaluation_epsilon = 0.05 self.num_heatup_steps = 50000 - self.evaluation_episodes = 25 - self.evaluate_every_x_episodes = 1000 + self.evaluation_episodes = 1 + self.evaluate_every_x_episodes = 100 class Breakout_DistributionalDQN(Preset):