Question about programming preference

1

when define a new net I find that I would use an object of a nn.Module class for multiple times so I just defined this object one time and use it in every nn.Module as follow:

feature = FeatureExtracter(SUBMODEL)
        
class Net1(nn.Module): 
        def forward(self,x):
            s_feats =feature(x,LAYER)
            ......#do something
            return ......

class Net2(nn.Module): 
        def forward(self,x):
            s_feats =feature(x,LAYER)
            ......#do something
            return ......

was that a bad programming habit ? or is there any reason in pytorch I shouldn’t do that, and instead define new object in every Net like this:

class Net1(nn.Module): 
        def __init__(self):
            super().__init__()
            self.feature = FeatureExtracter(SUBMODEL)
        def forward(self,x):
            s_feats = self.feature(x,LAYER)
            ......
            return ......

class Net2(nn.Module): 
        def __init__(self):
            super().__init__()
            self.feature = FeatureExtracter(SUBMODEL)
        def forward(self,x):
            s_feats = self.feature(x,LAYER)
            ......
            return ......
2

Both models are fine, neither is bad programming habit.