def test_naf_layer_diag():
batch_size = 2
for nb_actions in (1, 3):
# Construct single model with NAF as the only layer, hence it is fully deterministic
# since no weights are used, which would be randomly initialized.
L_flat_input = Input(shape=(nb_actions, ))
mu_input = Input(shape=(nb_actions, ))
action_input = Input(shape=(nb_actions, ))
x = NAFLayer(nb_actions,
mode='diag')([L_flat_input, mu_input, action_input])
model = Model(inputs=[L_flat_input, mu_input, action_input], outputs=x)
model.compile(loss='mse', optimizer='sgd')
# Create random test data.
L_flat = np.random.random((batch_size, nb_actions)).astype('float32')
mu = np.random.random((batch_size, nb_actions)).astype('float32')
action = np.random.random((batch_size, nb_actions)).astype('float32')
# Perform reference computations in numpy since these are much easier to verify.
P = np.zeros((batch_size, nb_actions, nb_actions)).astype('float32')
for p, l_flat in zip(P, L_flat):
p[np.diag_indices(nb_actions)] = l_flat
print(P, L_flat)
A_ref = np.array([
np.dot(np.dot(a - m, p), a - m) for a, m, p in zip(action, mu, P)
]).astype('float32')
A_ref *= -.5
# Finally, compute the output of the net, which should be identical to the previously
# computed reference.
A_net = model.predict([L_flat, mu, action]).flatten()
assert_allclose(A_net, A_ref, rtol=1e-5)
def test_naf_layer_full():
batch_size = 2
for nb_actions in (1, 3):
# Construct single model with NAF as the only layer, hence it is fully deterministic
# since no weights are used, which would be randomly initialized.
L_flat_input = Input(shape=((nb_actions * nb_actions + nb_actions) //
2, ))
mu_input = Input(shape=(nb_actions, ))
action_input = Input(shape=(nb_actions, ))
x = merge([L_flat_input, mu_input, action_input], mode='concat')
x = NAFLayer(nb_actions, mode='full')(x)
model = Model(input=[L_flat_input, mu_input, action_input], output=x)
model.compile(loss='mse', optimizer='sgd')
# Create random test data.
L_flat = np.random.random(
(batch_size,
(nb_actions * nb_actions + nb_actions) // 2)).astype('float32')
mu = np.random.random((batch_size, nb_actions)).astype('float32')
action = np.random.random((batch_size, nb_actions)).astype('float32')
# Perform reference computations in numpy since these are much easier to verify.
L = np.zeros((batch_size, nb_actions, nb_actions)).astype('float32')
LT = np.copy(L)
for l, l_T, l_flat in zip(L, LT, L_flat):
l[np.tril_indices(nb_actions)] = l_flat
l[np.diag_indices(nb_actions)] = np.exp(
l[np.diag_indices(nb_actions)])
l_T[:, :] = l.T
P = np.array([np.dot(l, l_T)
for l, l_T in zip(L, LT)]).astype('float32')
A_ref = np.array([
np.dot(np.dot(a - m, p), a - m) for a, m, p in zip(action, mu, P)
]).astype('float32')
A_ref *= -.5
# Finally, compute the output of the net, which should be identical to the previously
# computed reference.
A_net = model.predict([L_flat, mu, action]).flatten()
assert_allclose(A_net, A_ref, rtol=1e-5)