CODEWITHSUNDEEP
pythonpytorch
blue-sky
blue-sky

Reputation: 53786

Understanding PyTorch prediction

For my trained model this code : 

model(x[0].reshape(1,784).cuda())

returns : 

tensor([[-1.9903, -4.0458, -4.1143, -4.0074, -3.5510,  7.1074]], device='cuda:0')

My network model is defined as : 

# Hyper-parameters 
input_size = 784
hidden_size = 50
num_classes = 6
num_epochs = 5000
batch_size = 1
learning_rate = 0.0001

    criterion = nn.CrossEntropyLoss()
    optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)  

        class NeuralNet(nn.Module):
            def __init__(self, input_size, hidden_size, num_classes):
                super(NeuralNet, self).__init__()
                self.fc1 = nn.Linear(input_size, hidden_size) 
                self.relu = nn.ReLU()
                self.fc2 = nn.Linear(hidden_size, num_classes)  

            def forward(self, x):
                out = self.fc1(x)
                out = self.relu(out)
                out = self.fc2(out)
                return out

I'm attempting to understand the returned value : 

tensor([[-1.9903, -4.0458, -4.1143, -4.0074, -3.5510,  7.1074]], device='cuda:0')

The value 7.1074 is the most probable as it maximum value in tensor array ? As 7.1074 is at position 5, is the significance here that the associated output value being predicted for input x[0] is 5 ? If so what is the intuition behind this ?

Upvotes: 2

Views: 1241

Answers (1)

Zenith
Zenith

Reputation: 245

Disclaimer: I don't really know pytorch, but i'm guessing based on other libraries and general standard practice as i know it.

I believe it's the outputs of the last layer, which would be that fc2 linear transformation. So, the predicted category would be category 5, having the highest value.

You could think of it as being a sort of 'raw' probability distribution-esque range, that is commonly then squashed into the range (0,1] via softmax.

Upvotes: 2

Related Questions