Method __init__

Initializes the agent and constructs the Graph. Args: sess: A `tf.Session` object for running associated ops. num_actions: Int, number of actions the agent can take at any state. kappa: Float, Huber loss cutoff. network: tf.Keras.Model, expects 3 parameters: num_actions,

(self,
               sess,
               num_actions,
               kappa=1.0,
               network=atari_helpers.QuantileNetwork,
               num_atoms=200,
               gamma=0.99,
               update_horizon=1,
               min_replay_history=50000,
               update_period=4,
               target_update_period=10000,
               epsilon_fn=dqn_agent.linearly_decaying_epsilon,
               epsilon_train=0.1,
               epsilon_eval=0.05,
               epsilon_decay_period=1000000,
               replay_scheme='prioritized',
               tf_device='/cpu:0',
               optimizer=tf.train.AdamOptimizer(
                   learning_rate=0.00005, epsilon=0.0003125),
               summary_writer=None,
               summary_writing_frequency=500,
               minq_weight=10.0)