Priming

---

source

Priming_ECM

 Priming_ECM (num_actions:int, glow:float=0.1, damp:float=0.01,
              softmax:float=0.5, action_primes:list=None)

This sub-class of the Two-Layer ECM adds a variable for action priming. This variable should be a list of floats, each element of which corresponds to an action in the ECM. These “priming values” are summed with h-values of any edge connected to the associated action node prior to calculating walk probabilites with the softmax function

	Type	Default	Details
num_actions	int		The number of available actions.
glow	float	0.1	The glow (or eta) parameter.
damp	float	0.01	The damping (or gamma) parameter.
softmax	float	0.5	The softmax (or beta) parameter.
action_primes	list	None	weights on the probability that deliberation steps into each action. Defaults to 0 for each action

Example

test_ECM = Priming_ECM(num_actions = 2, action_primes = [0., 1.5])
#Number of steps to run simulation
T = 100
data_log = [None] * T
env = RLGL() #create a default red-light-green-light environment

for t in range(T):
    observation = env.get_observation()
    action = test_ECM.sample(observation)
    reward = env.get_reward(action)
    test_ECM.learn(reward)
    data_log[t] = {"env_state": env.state, "action": action, "reward": reward}
    env.transition(action)

plt.plot(range(T), [np.mean([data_log[step]["reward"] for step in range(i-10,i+1) if step >= 0]) for i in range(T)]) #plot a 10 step moving average of the reward