def test_mean_agg_constructor():
agg = MeanAggregator(2)
assert agg.output_dim == 2
assert not agg.has_bias
# Check config
config = agg.get_config()
assert config["output_dim"] == 2
assert config["bias"] is False
assert config["act"] == "relu"
def test_mean_agg_apply():
agg = MeanAggregator(5, bias=True, act=lambda x: x, kernel_initializer="ones")
inp1 = keras.Input(shape=(1, 2))
inp2 = keras.Input(shape=(1, 2, 2))
out = agg([inp1, inp2])
assert agg.weight_dims == [3, 2]
model = keras.Model(inputs=[inp1, inp2], outputs=out)
x1 = np.array([[[1, 1]]])
x2 = np.array([[[[2, 2], [3, 3]]]])
actual = model.predict([x1, x2])
expected = np.array([[[2, 2, 2, 5, 5]]])
assert expected == pytest.approx(actual)
def test_mean_agg_zero_neighbours():
agg = MeanAggregator(4, bias=False, act=lambda x: x, kernel_initializer="ones")
inp1 = keras.Input(shape=(1, 2))
inp2 = keras.Input(shape=(1, 0, 2))
out = agg([inp1, inp2])
model = keras.Model(inputs=[inp1, inp2], outputs=out)
x1 = np.array([[[1, 1]]])
x2 = np.zeros((1, 1, 0, 2))
actual = model.predict([x1, x2])
expected = np.array([[[2, 2, 2, 2]]])
assert expected == pytest.approx(actual)
def test_mean_agg_apply_groups():
agg = MeanAggregator(11, bias=True, act=lambda x: x, kernel_initializer="ones")
inp1 = keras.Input(shape=(1, 2))
inp2 = keras.Input(shape=(1, 2, 2))
inp3 = keras.Input(shape=(1, 2, 2))
out = agg([inp1, inp2, inp3])
assert agg.weight_dims == [5, 3, 3]
model = keras.Model(inputs=[inp1, inp2, inp3], outputs=out)
x1 = np.array([[[1, 1]]])
x2 = np.array([[[[2, 2], [3, 3]]]])
x3 = np.array([[[[5, 5], [4, 4]]]])
actual = model.predict([x1, x2, x3])
print(actual)
expected = np.array([[[2] * 5 + [5] * 3 + [9] * 3]])
assert expected == pytest.approx(actual)
def test_mean_agg_constructor_1():
agg = MeanAggregator(output_dim=4, bias=True, act=lambda x: x + 1)
assert agg.output_dim == 4
assert agg.has_bias
assert agg.act(2) == 3