class netCompressed( Net):
    def __init__(self, net, r):
        super().__init__()
        layer_weight = self.compressing('fc1.weight')
        self.fc1 = nn.Linear(layer_weight.shape[0],layer_weight.shape[1],bias = False)
        self.fc1.weight = layer_weight

        layer_weight = self.compressing('fc2.weight')
        self.fc2 = nn.Linear(layer_weight.shape[0],layer_weight.shape[1],bias = False)
        self.fc2.weight = layer_weight

        layer_weight = self.compressing('fc3.weight')
        self.fc3 = nn.Linear(layer_weight.shape[0],layer_weight.shape[1],bias = False)
        self.fc3.weight = layer_weight

        layer_weight = self.compressing('fc4.weight')
        self.fc4 = nn.Linear(layer_weight.shape[0],layer_weight.shape[1],bias = False)
        self.fc4.weight = layer_weight

        layer_weight = self.compressing('fc5.weight')
        self.fc5 = nn.Linear(layer_weight.shape[0],layer_weight.shape[1],bias = False)
        self.fc5.weight = layer_weight

        layer_weight = self.compressing('fc6.weight')
        self.fc6 = nn.Linear(layer_weight.shape[0],layer_weight.shape[1],bias = False)
        self.fc6.weight = layer_weight

        self.ReLu = torch.ReLu

    def forward(self, x):
        x = self.fc1(x.view(-1, 3 * 32*32))
        x = self.ReLU(x)
        x = self.fc2(x)
        x = self.ReLU(x)
        x = self.fc3(x)
        x = self.ReLU(x)
        x = self.fc4(x)
        x = self.ReLU(x)
        x = self.fc5(x)
        x = self.ReLU(x)
        x = self.fc6(x)
        return F.log_softmax(x, dim=1)

    def compressing(key):
        u,s,v  = torch.svd(net.state_dict()[key])
        main_s = s [:r]
        main_u = u[:r, : r]
        main_v = v[:r, :r]
        compressed_weight = torch.matmul(torch.matmul(main_u, torch.diag_embed(main_s)), main_v.transpose())
        return compressed_weight