def compute_entropy(logits):
"""
:param logits: A matrix of size N * |A|
:return: A vector of size N
"""
logp = log_softmax(logits)
return -np.sum(logp * np.exp(logp), axis=-1)
def compute_entropy(logits):
"""
:param logits: A matrix of size N * |A|
:return: A vector of size N
"""
logp = log_softmax(logits)
return -np.sum(logp * np.exp(logp), axis=-1)
def cartpole_get_logp_action(theta, ob, action):
"""
:param theta: A matrix of size |A| * (|S|+1)
:param ob: A vector of size |S|
:param action: An integer
:return: A scalar
"""
return log_softmax(compute_logits(theta, ob))[action]
def cartpole_get_logp_action(theta, ob, action):
"""
:param theta: A matrix of size |A| * (|S|+1)
:param ob: A vector of size |S|
:param action: An integer
:return: A scalar
"""
return log_softmax(compute_logits(theta, ob))[action]
