def _short_cut(self, name):
conv2d_ins = Conv2D(info="conv2d",
filters=self.config(self.KEYS.CONFIG.FILTERS),
kernel_size=(1, 1),
strides=(1, 1),
padding='same',
activation='basic')
return Stack(info=self.info.child_scope(name),
models=conv2d_ins,
nb_layers=2)
def test_stack_basic(clean_config):
models = [
Conv2D('conv1', 64, 3),
Conv2D('conv2', 128, 3),
Conv2D('conv3', 256, 3)
]
x = tf.ones([32, 64, 64, 3], dtype=tf.float32)
st = Stack(models)
res = st(x)
assert shape(res) == [32, 64, 64, 256]
def test_stack_parameters(clean_config):
models = [
Conv2D('conv1', 64, 3),
Conv2D('conv2', 128, 3),
Conv2D('conv3', 256, 3)
]
x = tf.ones([32, 64, 64, 3], dtype=tf.float32)
st = Stack(models)
res = st(x)
assert st.parameters[0].get_shape() == (3, 3, 3, 64)
def test_Stack(self):
x = self.get_input()
nb_layers = 2
y_ = x
stack_ins = Stack('Stack_test', tf.constant(x), self.make_model(),
nb_layers)
y = stack_ins()
with self.variables_initialized_test_session() as sess:
y = sess.run(y)
self.assertAllEqual(y, y_)
def fmap(self, m):
from dxl.learn.model.stack import Stack
return Stack([m, self])