Output of nn.Module

1

I put a simple model below

class NetA(nn.Module):
    def __init__(self):
        super(NetA, self).__init__()
        self.conv1 = nn.Conv2d(3, 6, (5, 5)) 
        self.conv2 = nn.Conv2d(6, 16, (5, 5))
        self.drop1 = nn.Dropout(p=0.5)
        self.fc1 = nn.Linear(16 * 5 * 5, 64) 
        self.fc2 = nn.Linear(64, 3) 
    def forward(self, x):
        x = F.max_pool2d(F.relu(self.conv1(x)), 2) 
        x = self.drop1(x)
        x = F.max_pool2d(F.relu(self.conv2(x)), 2) 
        x = x.view(-1, 16 * 5 * 5) 
        x = F.relu(self.fc1(x))

        if 1: # pattern1
            x = self.fc2(x)
        else: #pattern2
            x = F.softmax(self.fc2(x), dim=1)

        return x

If I input certain random value to the model, an output is [[0. 0. 0.]].
But, when activating the last line, and inputting certain random value, the output is, for instance, [[0.2243069 0.3878466 0.3878466]].

My question is why “x = F.softmax(self.fc2(x), dim=1)” gives me some values although x = self.fc2(x), the previous function in the model, gives exsactly 0,0,0 ??

2

Hi, can you please specify the exact input size you are working with?
Also, are you sure it’s exactly 0, 0, 0 and not some values that are displaying zero due to limited precision?

3

thank you for your comment,

the input is Input [input] shape: (1, 3, 32, 32)
with np.random.random(input_shape)

and output values are
[[0.000000000000000000000000
0.000000000000000000000000
0.000000000000000000000000]]

There was my wrong explanation.
The pattern1 is “x = self.fc2(x); return x”
This gives 0,0,0.

The pattern2 is “x = F.softmax(self.fc2(x), dim=1); return x”
This gives some values although self.fc2(x) gives 0,0,0