def __init__(self):
super().__init__()
self.sequntial = [
nn.linear(784, 300),
nn.ReLU(),
nn.linear(300, 300),
nn.ReLU(),
nn.linear(300, 10)
]
self.activation = []
def __init__(self):
super().__init__()
self.sequntial = [
nn.Conv2d(1, 16, 3),
nn.ReLU(),
nn.MaxPool2d(2),
nn.Conv2d(16, 32, 3),
nn.ReLU(),
nn.MaxPool2d(2),
nn.Conv2d(32, 64, 3),
nn.ReLU(),
nn.MaxPool2d(2),
nn.Conv2d(64, 128, 3),
nn.ReLU(),
nn.MaxPool2d(2),
nn.flatten(),
nn.linear(1 * 1 * 128, 10)
]
self.activation = []
def __init__(self):
super().__init__()
self.CNNSequential = SeqModule([
# convolution 1 + Relu
nn.Conv2d(1, 4, 3),
nn.ReLU(),
# convolution 2 + Relu + pooling
nn.Conv2d(4, 4, 3),
nn.ReLU(),
nn.MaxPool2d(2),
# convolution 3 + Relu + pooling
nn.Conv2d(4, 8, 3),
nn.ReLU(),
# convolution 4 + Relu + pooling
nn.Conv2d(8, 8, 3),
nn.ReLU(),
nn.MaxPool2d(2),
])
# 5층까지
self.CNN2Sequential = SeqModule([
nn.Conv2d(8, 16, 3),
nn.ReLU(),
])
############
# Concatenate
############
self.afterCNN = SeqModule([
nn.MaxPool2d(2, padding=1),
# same as reshape
nn.flatten(),
])
self.lastLayer = SeqModule(
[nn.linear(384, 128),
nn.ReLU(), nn.linear(128, 10)])
def __init__(self, embed, dropout=False):
super().__init__()
self.CNNSequential = SeqModule([
# convolution 1 + Relu
nn.Conv2d(3, 16, 3),
nn.ReLU(),
# convolution 2 + Relu + pooling
nn.Conv2d(16, 16, 3),
nn.ReLU(),
nn.MaxPool2d(2),
# convolution 3 + Relu + pooling
nn.Conv2d(16, 32, 3),
nn.ReLU(),
nn.MaxPool2d(2),
# convolution 4 + Relu + pooling
nn.Conv2d(32, 32, 3),
nn.ReLU(),
nn.MaxPool2d(2),
])
# 5층까지
self.CNN2Sequential = SeqModule([
nn.Conv2d(32, 64, 3),
nn.ReLU(),
])
############
# Concatenate
############
self.afterCNN = SeqModule([
nn.MaxPool2d(2),
# same as reshape
nn.flatten(),
nn.Dropout(check=dropout),
nn.linear(1536, 256),
nn.tanh()
])
self.Dropout = nn.Dropout(check=dropout)
self.RNN = SeqModule([
nn.RNN(embed, 256),
nn.tanh(),
])
############
# 중간에 ADD
############
self.lastLayer = SeqModule(
[nn.linear(256, 256),
nn.ReLU(), nn.linear(256, 1665)])