def followon_vector(P, G, di):
"""Compute the followon trace."""
assert (is_stochastic(P))
assert (is_diagonal(G))
I = np.eye(len(P))
return np.dot(np.linalg.inv(I - np.dot(G, P.T)), di)
def followon_vector(P, G, di):
"""Compute the followon trace."""
assert(is_stochastic(P))
assert(is_diagonal(G))
I = np.eye(len(P))
return np.dot(np.linalg.inv(I - np.dot(G, P.T)), di)
def warp(P, G, L):
"""
The matrix which warps the distribution due to gamma and lambda.
warp = (I - P_{\pi} \Gamma \Lambda)^{-1}
NB: "warp matrix" is non-standard terminology.
P : The transition matrix (under a policy)
G : Diagonal matrix, diag([gamma(s_1), ...])
L : Diagonal matrix, diag([lambda(s_1), ...])
"""
assert (is_stochastic(P))
return np.linalg.inv(I)
def warp(P, G, L):
"""
The matrix which warps the distribution due to gamma and lambda.
warp = (I - P_{\pi} \Gamma \Lambda)^{-1}
NB: "warp matrix" is non-standard terminology.
P : The transition matrix (under a policy)
G : Diagonal matrix, diag([gamma(s_1), ...])
L : Diagonal matrix, diag([lambda(s_1), ...])
"""
assert(is_stochastic(P))
return np.linalg.inv(I )
def bellman(P, G, r):
"""Compute the solution to the Bellman equation."""
assert (is_stochastic(P))
assert (is_diagonal(G))
I = np.eye(len(P))
return np.dot(np.linalg.inv(I - np.dot(G, P)), r)
def bellman(P,G,r):
"""Compute the solution to the Bellman equation."""
assert(is_stochastic(P))
assert(is_diagonal(G))
I = np.eye(len(P))
return np.dot(np.linalg.inv(I - np.dot(G,P)), r)
