In the following code, a Deep Q-learning solution for the FrozenLake problem is proposed:
import gymimport numpy as npfrom keras.models import Sequentialfrom keras.layers.core import Dense, Reshapefrom keras.layers.embeddings import Embeddingfrom keras.optimizers import Adamfrom rl.agents.dqn import DQNAgentfrom rl.policy import BoltzmannQPolicyfrom rl.memory import SequentialMemory ENV_NAME = 'FrozenLake-v0' env = gym.make(ENV_NAME)np.random.seed(1)env.seed(1)Actions = env.action_space.n model = Sequential()model.add(Embedding(16, 4, input_length=1))model.add(Reshape((4,)))print(model.summary()) memory = SequentialMemory(limit=10000, window_length=1)policy = BoltzmannQPolicy()Dqn = DQNAgent(model=model, nb_actions=Actions, ...
