recurrentLSTMNetwork.py 文件源码

def forward(self, x, prevState = None ):

        # dimensions
        if len(x.size()) == 2: x = x.unsqueeze(0)
        batch = x.size(0)
        steps = x.size(1)

        if prevState == None: prevState = {}
        hs = {}
        cs = {}
        for t in range(steps):
            # xt
            xt = x[:,t,:]
            # prev h and pre c
            hp = hs[t-1] or prevState.h or torch.zeros()
        a = 0