def test_prop_up(self):
res = self.rbm.prop_up(self.visible_layer)
expected = neur.sigmoid(np.array([[ 1.260987, 2.484167],
[ 1.142713, 1.844246],
[ 0.878157, 1.070012],
[ 0.364556, 0.974234]]))
self.assertTrue(self.all_equalish(res, expected))
def encode(self, x):
# add ones
m = len(x)
x = np.hstack([np.ones((m,1)), x])
return neur.sigmoid(np.dot(x, self.encode_weights.transpose()))
def test_prop_up(self):
res = self.rbm.prop_up(self.visible_layer)
expected = neur.sigmoid(
np.array([[1.260987, 2.484167], [1.142713, 1.844246],
[0.878157, 1.070012], [0.364556, 0.974234]]))
self.assertTrue(self.all_equalish(res, expected))
def prop_up(self, vis_layer):
return neur.sigmoid(np.dot(vis_layer, self.weights.transpose()))
def prop_down(self, hid_layer):
return neur.sigmoid(np.dot(hid_layer, self.weights))
def test_sigmoid(self):
res = neur.sigmoid(np.array(([1, 2], [3, 4])))
target = np.array([[0.7310585, 0.88079708], [0.95257413, 0.98201379]])
self.equalish(res, target)
def prop_down(self, hid_layer):
vis_bias = self.expand_visible_bias(hid_layer.shape[0])
return neur.sigmoid(np.dot(hid_layer, self.weights) + vis_bias)
def prop_up(self, vis_layer):
hid_bias = self.expand_hidden_bias(vis_layer.shape[0])
res = np.dot(vis_layer, self.weights.transpose()) + hid_bias
return neur.sigmoid(res)
def test_sigmoid(self):
res = neur.sigmoid(np.array(([1, 2], [3, 4])))
target = np.array([[0.7310585, 0.88079708], [0.95257413, 0.98201379]])
self.equalish(res, target)
def prop_down(self, hid_layer):
vis_bias = self.expand_visible_bias(hid_layer.shape[0])
return neur.sigmoid(np.dot(hid_layer, self.weights) + vis_bias)
def prop_up(self, vis_layer):
hid_bias = self.expand_hidden_bias(vis_layer.shape[0])
res = np.dot(vis_layer, self.weights.transpose()) + hid_bias
return neur.sigmoid(res)
