Pytorch解决LSTM前向传播时报错：RuntimeError: input must have 3 dimensions, got 2

完整报错为：RuntimeError: input must have 3 dimensions, got 2

解决方法

前向传播代码中，加入：

def forward(self, x)
	x = x.view(len(x), 1, -1) # 把原有2维度[a,b]改为3维[a,1,b]
	# ......后续的代码

问题解析

比如我定义的LSTM：

class LSTM(nn.Module):
    def __init__(self, input_size=5, hidden_layer_size=100, output_size=1):
		# ....初始化
		pass
    def forward(self, input_x): # input_x.size() 为 [20个数据,5个维度]
        lstm_out, self.hidden_cell = self.lstm(input_x, self.hidden_cell)
        predictions = self.linear(lstm_out.view(len(input_x), -1))
        return predictions[-1]

就会报错，改为如下即正确：

class LSTM(nn.Module):
    def __init__(self, input_size=5, hidden_layer_size=100, output_size=1):
		# ....初始化
		pass
    def forward(self, input_x): # input_x.size() 为 [20个数据,5个维度]
        input_x = input_x.view(len(input_x), 1, -1) # 维度变为 [20,1,5]
        lstm_out, self.hidden_cell = self.lstm(input_x, self.hidden_cell)
        predictions = self.linear(lstm_out.view(len(input_x), -1))
        return predictions[-1]