hw2笔记

列表推导式

self.fc = nn.Sequential(
    BasicBlock(input_dim, hidden_dim),
    # 循环实现N次，1024*1024全连接层 
    *[BasicBlock(hidden_dim, hidden_dim) for _ in range(hidden_layers)],
    nn.Linear(hidden_dim, output_dim)
)

repeat()

https://blog.csdn.net/m0_46412065/article/details/128043821

view()

N = 420,L = 39 * 21,H_in = 39
x = torch.randn(N,L)
x.view(N,-1,H_in) # (420,21,39)

permute()

https://www.jb51.net/article/246243.htm

BatchNorm1D/BatchNorm2D

# BatchNorm1D 适用于2D或3D输入,即 (N,C) 或 (N,C,L)

# BatchNorm1D 适用于4D输入,即 (N,C,H,W)

RNN/LSTM/GRU

# RNN 1982 
model_base = nn.RNN(
    input_size= 4,
    hidden_size= 256,
    #循环神经网络的层数
    num_layers= 1,
    #方向神经网络
    bidirectional= False
)

x = torch.randn(size=(365,8,4))

# L:对数据的定量长度，如小时、天、月等
# N:批量大小
# H_in:单个时间步数据的长度
output, h_n = model_base(x)


# LSTM 1997
model_lstm = nn.LSTM(
    input_size=4,
    hidden_size=256,
    num_layers=1,
    bidirectional=False
)
output, (h_n, c_n) = model_lstm(x)


# GRU 2014
model_gru = nn.GRU(
    input_size= 4,
    hidden_size= 256,
    num_layers= 1,
    bidirectional= False
)

output, h_n = model_gru(x)

模型参数计算

print("base's parameters : {}".format( sum([p.numel() for p in model_base.parameters() if p.requires_grad])))