class Test_GaussianBinaryVarianceRBM(unittest.TestCase):
gbrbmData = numx.array([[0.57459607, 0.57689834],
[-0.60788602, -0.57941004]])
gbrbmw = numx.array([[0.12179488, 2.95950177], [
0.20318085,
-28.62372894,
]])
gbrbmbv = numx.array([[-19.24399659, -13.26258696]])
gbrbmbh = numx.array([[-0.11155958, 57.02097584]])
gbrbm = Model.GaussianBinaryVarianceRBM(2, 2, gbrbmData, gbrbmw, gbrbmbv,
gbrbmbh, 1.0, 0.0, 0.0)
gbrbmTruelogZ = 59.6749019726
gbrbmTrueLL = -1.7328699078
gbrbmBestLLPossible = -1.732867951
epsilon = 0.00001
def test___init__(self):
sys.stdout.write(
'GaussianBinaryVarianceRBM -> Performing init test ...')
sys.stdout.flush()
print(' successfully passed!')
sys.stdout.flush()
pass
def test__calculate_sigma_gradient(self):
sys.stdout.write(
'GaussianBinaryVarianceRBM -> Performing calculate_sigma_gradient test ...'
)
sys.stdout.flush()
deltaSigma = self.gbrbm._calculate_sigma_gradient(
numx.array([[0.98, -0.56], [-0.3, 0.8]]),
numx.array([[0, 1], [1, 1]]))
target = numx.array([[763.9331994, 372.52636802]])
assert numx.all(numx.abs(target - deltaSigma) < self.epsilon)
print(' successfully passed!')
sys.stdout.flush()
def test_get_parameters(self):
sys.stdout.write(
'GaussianBinaryVarianceRBM -> Performing get_parameters test ...')
sys.stdout.flush()
assert len(self.gbrbm.get_parameters()) == 4
print(' successfully passed!')
sys.stdout.flush()
def test_calculate_gradients(self):
sys.stdout.write(
'GaussianBinaryVarianceRBM -> Performing calculate_gradients test ...'
)
sys.stdout.flush()
assert len(
self.gbrbm.calculate_gradients(
numx.array([[0.98, -0.56], [-0.3, 0.8]]),
numx.array([[0, 1], [1, 1]]))) == 4
print(' successfully passed!')
sys.stdout.flush()
# Set restriction factor, learning rate, batch size and maximal number of epochs
restrict = 0.01 * numx.max(numxext.get_norms(train_data, axis=1))
eps = 0.1
batch_size = 100
max_epochs = 200
# Create model, initial weights=Glorot init., initial sigma=1.0, initial bias=0,
# no centering (Usually pass the data=training_data for a automatic init. that is
# set the bias and sigma to the data mean and data std. respectively, for
# whitened data centering is not an advantage)
rbm = model.GaussianBinaryVarianceRBM(number_visibles=v1 * v2,
number_hiddens=h1 * h2,
initial_weights='AUTO',
initial_visible_bias=0,
initial_hidden_bias=0,
initial_sigma=1.0,
initial_visible_offsets=0.0,
initial_hidden_offsets=0.0,
dtype=numx.float64)
# Set the hidden bias such that the scaling factor is 0.01
rbm.bh = -(numxext.get_norms(rbm.w + rbm.bv.T, axis=0) -
numxext.get_norms(rbm.bv, axis=None)) / 2.0 + numx.log(0.01)
rbm.bh = rbm.bh.reshape(1, h1 * h2)
# Training with CD-1
k = 1
trainer_cd = trainer.CD(rbm)
# Train model, status every 10th epoch