model.py

from torch import nn
import torch.nn.functional as F
import torch


class activation():

    def __init__(self, act_type, negative_slope=0.2, inplace=True):
        super().__init__()
        self._act_type = act_type
        self.negative_slope = negative_slope
        self.inplace = inplace

    def __call__(self, input):
        if self._act_type == 'leaky':
            return F.leaky_relu(input, negative_slope=self.negative_slope, inplace=self.inplace)
        elif self._act_type == 'relu':
            return F.relu(input, inplace=self.inplace)
        elif self._act_type == 'sigmoid':
            return torch.sigmoid(input)
        else:
            raise NotImplementedError


class ED(nn.Module):

    def __init__(self, encoder, decoder):
        super().__init__()
        self.encoder = encoder
        self.decoder = decoder

    def forward(self, input):
        state = self.encoder(input) #encode-->hidden state
        output = self.decoder(state)#hidden state--》decode
        return output