Accessing different layers in pytorch

I want to access my network last layer in a class but I got this error!

import torch
import torch.nn as nn
import numpy as np

class Net(nn.Module):
def init(self, number_of_class, ):
super(Net, self).init()
self._conv1 = nn.Conv2d(1, 32, kernel_size=(3, 5))
self._pool1 = nn.MaxPool2d(kernel_size=(1, 3))
self._batch_norm1 = nn.BatchNorm2d(32)
self._prelu1 = nn.PReLU(32)
self._dropout1 = nn.Dropout2d(.5)

    self._conv2 = nn.Conv2d(32, 64, kernel_size=(3, 5))
    self._pool2 = nn.MaxPool2d(kernel_size=(1, 3))
    self._batch_norm2 = nn.BatchNorm2d(64)
    self._prelu2 = nn.PReLU(64)
    self._dropout2 = nn.Dropout2d(.5)
    self._fc1 = nn.Linear(3072, 500)
    self._batch_norm3 = nn.BatchNorm1d(500)
    self._prelu3 = nn.PReLU(500)
    self._dropout3 = nn.Dropout(.5)
    self.fc = nn.Linear(500, number_of_class)
    print("Number Parameters: ", self.get_n_params())

def get_n_params(self):
    model_parameters = filter(lambda p: p.requires_grad, self.parameters())
    number_params = sum([ for p in model_parameters])
    return number_params

def init_weights(self):
    for m in self.modules():

def initialize_weights(self):
    for m in self.modules():
        if isinstance(m, nn.Conv2d):
        elif isinstance(m, nn.Linear):

def forward(self, x):
    conv1 = self._dropout1(self._prelu1(self._batch_norm1(self._conv1(x))))
    # print(conv1.shape)
    pool1 = self._pool1(conv1)
    conv2 = self._dropout2(self._prelu2(self._batch_norm2(self._conv2(pool1))))
    # print(conv2.shape)
    pool2 = self._pool2(conv2)
    # print(pool2.shape)
    flatten_tensor = pool2.view(pool2.size(0), -1)
    # print(flatten_tensor.shape)
    fc1 = self._dropout3(self._prelu3(self._batch_norm3(self._fc1(flatten_tensor))))
    # print(fc1.shape)
    Output = self.fc(fc1)
    # print(Output.shape)
    return Output

in_features = Net.fc.in_features

type object ‘Net’ has no attribute ‘fcPreformatted text’

I had to something like this some time ago, and I just returned the output before the linear layer i.e.

before_fc1 = fc1
output = self.fc(fc1)
return before_fc1, output 

I am guessing you have to send this feature to your loss function, in that case this may work.