def __init__(self, D_in, H1, H2, D_out, weights=''):
self.FC1 = nn_layer.FC(D_in, H1)
self.ReLU1 = activation.ReLU()
self.FC2 = nn_layer.FC(H1, H2)
self.ReLU2 = activation.ReLU()
self.FC3 = nn_layer.FC(H2, D_out)
if weights == '':
pass
else:
# Load weights from file
with open(weights, 'rb') as f:
params = pickle.load(f)
self.set_params(params)
def __init__(self, D_in, H, D_out, weights=''):
'''
D_in: input feature dimension
H: number of hidden neurons, output dimension of first FC layer and input dimension of second FC layer
D_out: output dimension, which is 10 for digit recognition.
'''
self.FC1 = nn_layer.FC(D_in, H)
self.ReLU1 = activation.ReLU()
self.FC2 = nn_layer.FC(H, D_out)
if weights == '':
pass
else:
# Load weights from file
with open(weights, 'rb') as f:
params = pickle.load(f)
self.set_params(params)
def __init__(self):
self.Cin = 1
self.D_out = 10
# Cin: input channel
# Cout: output channel
# F: kernel size 3x3
# Conv1: Cin=1, Cout=6, F=3
self.conv1 = conv_layer.Conv (self.Cin, 6, 3)
self.ReLU1 = activation.ReLU ()
self.pool1 = pooling.MaxPool (2,2)
# Conv2: Cin=6, Cout=16, F=3
self.conv2 = conv_layer.Conv (6, 16, 3)
self.ReLU2 = activation.ReLU ()
self.pool2 = pooling.MaxPool (2,2)
# FC1 flatten to be 64*784
self.FC1 = nn_layer.FC(784, 120)
self.ReLU3 = activation.ReLU ()
self.FC2 = nn_layer.FC (120, 84)
self.ReLU4 = activation.ReLU ()
self.FC3 = nn_layer.FC (84, self.D_out)
self.Softmax = activation.Softmax ()
self.p2_shape = None
def __init__(self, D_in, H1, H2, D_out, weights=''):
self.FC1 = nn_layer.FC(D_in, H1)
self.ReLU1 = activation.ReLU()
self.FC2 = nn_layer.FC(H1, H2)
self.ReLU2 = activation.ReLU()
self.FC3 = nn_layer.FC(H2, D_out)