[TIP / Pytorch] Linear NN Model base Architecture

pytorch가 arhcitecture가 저장이 안 되니, 본 판때기를 잘 만들어서 한 구조에서 여러개의 파리미터를 넣을 수 있도록 해야 한다.

여기서는 본 판때기에 대한 base를 만들어봄.

 

from torch import nn

class Net(nn.Module) :
    def __init__(self, layers , activation,  bn, dropout) :
        super(Net,self).__init__()
        self.model = self.make_model(layers , activation,  bn, dropout)
        
    def forward(self, x) :
        self.model(x)
        
    def make_model(self, layers , activation,  bn, dropout) :
        model = []
        layers.append(1)
        if activation.lower() == "selu" :
            selu = nn.SELU
        for idx , layer in enumerate(layers[1:]) :
            mod = nn.Linear(layers[idx], layer)
            model.append(mod)
            if (idx+1) == len(layers[1:]) :
                pass
            else :
                if dropout == True :
                    model.append(nn.AlphaDropout(0.8))
                if bn == True :
                    model.append(nn.BatchNorm1d(layer))
                model.append(selu())
        return nn.Sequential(*model)

input_size = 150
layers = [input_size, 10,5,10]
layers = [input_size, 100,100,100,100,5,10]
activation = "selu"
dropout =True
bn = True
Net(layers,activation,bn,dropout)

Net(
  (model): Sequential(
    (0): Linear(in_features=150, out_features=10, bias=True)
    (1): AlphaDropout(p=0.8, inplace=False)
    (2): BatchNorm1d(10, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (3): SELU()
    (4): Linear(in_features=10, out_features=5, bias=True)
    (5): AlphaDropout(p=0.8, inplace=False)
    (6): BatchNorm1d(5, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (7): SELU()
    (8): Linear(in_features=5, out_features=10, bias=True)
    (9): AlphaDropout(p=0.8, inplace=False)
    (10): BatchNorm1d(10, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (11): SELU()
    (12): Linear(in_features=10, out_features=1, bias=True)
  )
)