def model(self):
input_tensor = Input(shape=self._input_shape, name='input')
flattener = Flatten()(input_tensor)
output = PolyFit(deg=self.num_hidden,
use_xbias=True,
name='output',
kernel_regularizer=regularizers.l2(
self.l2))(flattener)
model = Model(inputs=input_tensor, outputs=output)
return model
def test_PolyFit(self):
print('==========PolyFit tests==========')
from astroNN.nn.layers import PolyFit
# Data preparation
polynomial_coefficient = [0.1, -0.05]
random_xdata = np.random.normal(0, 3, (100, 1))
random_ydata = polynomial_coefficient[
1] * random_xdata + polynomial_coefficient[0]
self.assertRaises(ValueError,
PolyFit,
deg=1,
use_xbias=False,
init_w=[2., 3., 4.])
input = Input(shape=[
1,
])
output = PolyFit(deg=1,
use_xbias=False,
init_w=[[[0.1]], [[-0.05]]],
name='polyfit')(input)
model = Model(inputs=input, outputs=output)
model.compile(optimizer='sgd', loss='mse')
model.fit(random_xdata, random_ydata, batch_size=32, epochs=1)
# no gradient update thus the answer should equal to the real equation anyway
self.assertEqual(
np.allclose(model.predict(random_xdata), random_ydata), True)
# no gradients update because initial weights are perfect
npt.assert_almost_equal(np.squeeze(model.get_weights()[0]),
polynomial_coefficient)
print(model.get_layer('polyfit').get_config())
# Data preparation
polynomial_coefficient = [0.075, -0.05]
random_xdata = np.random.normal(0, 1, (100, 2))
random_ydata = np.vstack(
(np.sum(polynomial_coefficient[1] * random_xdata +
polynomial_coefficient[0],
axis=1),
np.sum(polynomial_coefficient[1] * random_xdata +
polynomial_coefficient[0],
axis=1))).T
input = Input(shape=[
2,
])
output = PolyFit(deg=1,
output_units=2,
use_xbias=False,
init_w=[[[0.075, 0.075], [0.075, 0.075]],
[[-0.05, -0.05], [-0.05, -0.05]]],
name='PolyFit')(input)
model = Model(inputs=input, outputs=output)
model.compile(optimizer='sgd', loss='mse')
model.fit(random_xdata, random_ydata, batch_size=32, epochs=1)
npt.assert_almost_equal(model.predict(random_xdata), random_ydata)
# no gradients update because initial weights are perfect
npt.assert_almost_equal(np.squeeze(model.get_weights()[0]),
[[[0.075, 0.075], [0.075, 0.075]],
[[-0.05, -0.05], [-0.05, -0.05]]])