def create_layers(self, state, **kwargs):
action = Variable(np.ones(env.n_actions))
output = SoftMax(in_layers=[
Reshape(in_layers=[action], shape=(-1, env.n_actions))
])
value = Variable([0.0])
return {'action_prob': output, 'value': value}
def test_variable(self):
"""Test that Variable can be invoked."""
value = np.random.uniform(size=(2, 3)).astype(np.float32)
with self.session() as sess:
out_tensor = Variable(value)()
sess.run(tf.global_variables_initializer())
assert np.array_equal(value, out_tensor.eval())
def test_variable(self):
"""Test that Variable can be invoked."""
value = np.random.uniform(size=(2, 3)).astype(np.float32)
with self.session() as sess:
out_tensor = Variable(value)()
sess.run(tf.global_variables_initializer())
assert np.array_equal(value, out_tensor.eval())
def test_convert_to_tensor(self):
"""Test implicit conversion of Layers to Tensors."""
v = Variable(np.array([1.5]))
v.create_tensor()
with self.session() as sess:
sess.run(tf.global_variables_initializer())
result = sess.run(v)
assert result == 1.5
result = sess.run(tf.gradients(v, v))
assert result[0] == 1.0
def test_convert_to_tensor(self):
"""Test implicit conversion of Layers to Tensors."""
v = Variable(np.array([1.5]))
v.create_tensor()
with self.session() as sess:
sess.run(tf.global_variables_initializer())
result = sess.run(v)
assert result == 1.5
result = sess.run(tf.gradients(v, v))
assert result[0] == 1.0
def test_initialize_variable(self):
"""Test methods for initializing a variable."""
tg = dc.models.TensorGraph(use_queue=False)
features = Feature(shape=(None, 1))
tg.set_loss(Dense(1, in_layers=features))
var = Variable([10.0])
tg.add_output(var)
tg.fit_generator([])
assert tg.predict_on_batch(np.zeros((1, 1))) == [10.0]
var.set_variable_initial_values([[15.0]])
tg.fit_generator([])
assert tg.predict_on_batch(np.zeros((1, 1))) == [15.0]
def test_submodels(self):
"""Test optimizing submodels."""
tg = dc.models.TensorGraph(learning_rate=0.1,
batch_size=1,
use_queue=False)
features = Feature(shape=(None, 1))
var1 = Variable([2.0])
var2 = Variable([2.0])
tg.add_output(var1)
tg.add_output(var2)
loss = (var1 - 1) * (var1 - 1) + (var2 - 1) * (var2 - 1) + features
tg.set_loss(loss)
subloss1 = var1 * var1 + features
subloss2 = var1 * var1 + var2 * var2 + features
submodel1 = tg.create_submodel(loss=subloss1)
submodel2 = tg.create_submodel(layers=[var2], loss=subloss2)
data = np.zeros((1, 1))
generator = [{features: data}] * 500
# Optimize submodel 1. This should send var1 to 0 while leaving var2 unchanged.
tg.fit_generator(generator, submodel=submodel1)
self.assertAlmostEqual(0.0,
tg.predict_on_batch(data, outputs=var1)[0],
places=4)
self.assertAlmostEqual(2.0,
tg.predict_on_batch(data, outputs=var2)[0],
places=4)
# Optimize the main loss. This should send both variables toward 1.
tg.fit_generator(generator)
self.assertAlmostEqual(1.0,
tg.predict_on_batch(data, outputs=var1)[0],
places=4)
self.assertAlmostEqual(1.0,
tg.predict_on_batch(data, outputs=var2)[0],
places=4)
# Optimize submodel 2. This should send var2 to 0 while leaving var1 unchanged.
tg.fit_generator(generator, submodel=submodel2)
self.assertAlmostEqual(1.0,
tg.predict_on_batch(data, outputs=var1)[0],
places=4)
self.assertAlmostEqual(0.0,
tg.predict_on_batch(data, outputs=var2)[0],
places=4)
def create_layers(self, state, **kwargs):
reshaped = Reshape(shape=(1, -1, 10), in_layers=state)
gru = GRU(n_hidden=10, batch_size=1, in_layers=reshaped)
output = SoftMax(
in_layers=[Reshape(in_layers=[gru], shape=(-1, env.n_actions))])
value = Variable([0.0])
return {'action_prob': output, 'value': value}
def test_Variable_pickle(): tg = TensorGraph() feature = Feature(shape=(tg.batch_size, 1)) layer = Variable(np.array([15.0])) output = Multiply(in_layers=[feature, layer]) tg.add_output(output) tg.set_loss(output) tg.build() tg.save()
def test_initialize_variable(self):
"""Test methods for initializing a variable."""
# Set by variable constructor.
tg = dc.models.TensorGraph(use_queue=False)
features = Feature(shape=(None, 1))
tg.set_loss(Dense(1, in_layers=features))
var = Variable([10.0])
tg.add_output(var)
assert tg.predict_on_batch(np.zeros((1, 1))) == [10.0]
# Set by set_variable_initial_values().
tg = dc.models.TensorGraph(use_queue=False)
features = Feature(shape=(None, 1))
tg.set_loss(Dense(1, in_layers=features))
var = Variable([10.0])
var.set_variable_initial_values([[15.0]])
tg.add_output(var)
assert tg.predict_on_batch(np.zeros((1, 1))) == [15.0]
def test_get_layer_variable_values(self):
"""Test to get the variable values associated with a layer"""
# Test for correct value return (normal mode)
tg = dc.models.TensorGraph()
var = Variable([10.0, 12.0])
tg.add_output(var)
expected = [10.0, 12.0]
obtained = tg.get_layer_variable_values(var)[0]
np.testing.assert_array_equal(expected, obtained)
# Test for shapes (normal mode)
tg = dc.models.TensorGraph()
input_tensor = Input(shape=(10, 100))
output = Dense(out_channels=20, in_layers=[input_tensor])
tg.add_output(output)
expected_shape = (100, 20)
obtained_shape = tg.get_layer_variable_values(output)[0].shape
assert expected_shape == obtained_shape