def __init__(self, settings, net_settings=None):
"""
:param settings: Full connected layer settings
:type settings: FullConnectedLayerSettings
"""
super(_FullConnectedLayer, self).__init__(settings, net_settings)
self.relu = _ReluLayer(ReluLayerSettings(in_shape=self.settings.out_shape, activation=self.settings.activation))
def test_forward(self):
arr = np.random.uniform(-5, 5, 48).reshape((4, 4, 3))
s = ReluLayerSettings(in_shape=arr.shape, activation='max')
l = _ReluLayer(s)
res = l.forward(arr)
print(arr[:, :, 0])
print(arr[:, :, 1])
print(arr[:, :, 2])
print('-----------')
print(res[:, :, 0])
print(res[:, :, 1])
print(res[:, :, 2])
def test_backward(self):
arr = np.random.uniform(-5, 5, 48).reshape((4, 4, 3))
s = ReluLayerSettings(in_shape=arr.shape, activation='max')
l = _ReluLayer(s)
l.prev_layer = _InputLayer(InputLayerSettings(in_shape=arr.shape))
l.next_layer = object()
e = np.random.uniform(-5, 5, 48).reshape(*s.out_shape)
print(e[:, :, 0])
print(e[:, :, 1])
print(e[:, :, 2])
print('-----------')
res = l.prev_layer.forward(arr)
res = l.forward(res)
res = l.backward(e)
print(res[:, :, 0])
print(res[:, :, 1])
print(res[:, :, 2])
