def log_prob(w, D):
agent = RWSoftmaxAgent(task=MyBanditTask(),
learning_rate=sigmoid(w[0]),
inverse_softmax_temp=stable_exp(w[1]))
for t in range(D.shape[0]):
x = D[t, :7]
u = D[t, 7:11]
r = D[t, 11]
x_ = D[t, 12:]
agent.log_prob(x, u)
agent.learning(x, u, r, x_, None)
J = np.diag([grad.sigmoid(w[0]), grad.exp(w[1])])
return -agent.logprob_, -J @ agent.grad_
def log_prob(w, D):
lr = sigmoid(w[0], a_min=-6, a_max=6)
ist = stable_exp(w[1], a_min=-10, a_max=10)
agent = RWSoftmaxAgent(TwoArmedBandit(), lr, ist)
L = 0
for t in range(D.shape[0]):
x = D[t, :3]
u = D[t, 3:5]
r = D[t, 5]
x_ = D[t, 6:]
agent.log_prob(x, u)
agent.learning(x, u, r, x_, None)
J = np.array([grad.sigmoid(w[0]), grad.exp(w[1])])
return -agent.logprob_, -J * agent.grad_,
def reparam_jac_rwssm(x):
return np.diag(np.array([grad.sigmoid(x[0]), grad.exp(x[1]), x[2]]))
def test_sigmoid():
x = np.linspace(-5, 5, 10)
f = lambda x: utils.sigmoid(x)
ag = elementwise_grad(f)(x)
fg = grad.sigmoid(x)
assert (np.all(np.linalg.norm(ag - fg) < 1e-6))