def test_log_likelihood_sym(self, output_dim, input_shape):
bmr = BernoulliMLPRegressorWithModel(input_shape=(input_shape[1], ),
output_dim=output_dim)
new_xs_var = tf.placeholder(tf.float32, input_shape)
new_ys_var = tf.placeholder(dtype=tf.float32,
name='ys',
shape=(None, output_dim))
data = np.full(input_shape, 0.5)
one_hot_label = np.zeros((input_shape[0], output_dim))
one_hot_label[np.arange(input_shape[0]), 0] = 1
p = bmr._f_prob(np.asarray(data))
ll = bmr._dist.log_likelihood(np.asarray(one_hot_label), dict(p=p))
outputs = bmr.log_likelihood_sym(new_xs_var, new_ys_var, name='ll_sym')
ll_from_sym = self.sess.run(outputs,
feed_dict={
new_xs_var: data,
new_ys_var: one_hot_label
})
assert np.allclose(ll, ll_from_sym, rtol=0, atol=1e-5)
def test_fit_normalized(self, input_shape, output_dim):
bmr = BernoulliMLPRegressorWithModel(input_shape=input_shape,
output_dim=output_dim)
observations, returns = get_train_data(input_shape, output_dim)
for _ in range(150):
bmr.fit(observations, returns)
paths, expected = get_test_data(input_shape, output_dim)
prediction = np.cast['int'](bmr.predict(paths['observations']))
assert np.allclose(prediction, expected, rtol=0, atol=0.1)
x_mean = self.sess.run(bmr.model.networks['default'].x_mean)
x_mean_expected = np.mean(observations, axis=0, keepdims=True)
x_std = self.sess.run(bmr.model.networks['default'].x_std)
x_std_expected = np.std(observations, axis=0, keepdims=True)
assert np.allclose(x_mean, x_mean_expected)
assert np.allclose(x_std, x_std_expected)
def test_is_pickleable2(self):
bmr = BernoulliMLPRegressorWithModel(input_shape=(1, ), output_dim=2)
with tf.variable_scope(
'BernoulliMLPRegressorWithModel/NormalizedInputMLPModel',
reuse=True):
x_mean = tf.get_variable('normalized_vars/x_mean')
x_mean.load(tf.ones_like(x_mean).eval())
x1 = bmr.model.networks['default'].x_mean.eval()
h = pickle.dumps(bmr)
with tf.Session(graph=tf.Graph()):
bmr_pickled = pickle.loads(h)
x2 = bmr_pickled.model.networks['default'].x_mean.eval()
assert np.array_equal(x1, x2)
def test_is_pickleable(self):
bmr = BernoulliMLPRegressorWithModel(input_shape=(1, ), output_dim=2)
with tf.variable_scope(
'BernoulliMLPRegressorWithModel/NormalizedInputMLPModel',
reuse=True):
bias = tf.get_variable('mlp/hidden_0/bias')
bias.load(tf.ones_like(bias).eval())
bias1 = bias.eval()
result1 = np.cast['int'](bmr.predict(np.ones((1, 1))))
h = pickle.dumps(bmr)
with tf.Session(graph=tf.Graph()):
bmr_pickled = pickle.loads(h)
result2 = np.cast['int'](bmr_pickled.predict(np.ones((1, 1))))
assert np.array_equal(result1, result2)
with tf.variable_scope(
'BernoulliMLPRegressorWithModel/NormalizedInputMLPModel',
reuse=True):
bias2 = tf.get_variable('mlp/hidden_0/bias').eval()
assert np.array_equal(bias1, bias2)
def test_optimizer_args(self, mock_cg, mock_lbfgs):
lbfgs_args = dict(max_opt_itr=25)
cg_args = dict(cg_iters=15)
bmr = BernoulliMLPRegressorWithModel(
input_shape=(1, ),
output_dim=2,
optimizer=LbfgsOptimizer,
optimizer_args=lbfgs_args,
tr_optimizer=ConjugateGradientOptimizer,
tr_optimizer_args=cg_args,
use_trust_region=True)
assert mock_lbfgs.return_value is bmr._optimizer
assert mock_cg.return_value is bmr._tr_optimizer
mock_lbfgs.assert_called_with(max_opt_itr=25)
mock_cg.assert_called_with(cg_iters=15)