python-pytorchHow do I calculate cross entropy loss using Python and PyTorch?

Cross entropy loss is a measure of how well a set of predicted probabilities match the true labels of a data set. It can be calculated using Python and PyTorch by first defining the true labels and predicted probabilities as tensors.

import torch

true_labels = torch.tensor([1, 0, 0, 1])
predicted_probabilities = torch.tensor([0.9, 0.2, 0.1, 0.8])

Then, the cross entropy loss can be calculated by using the torch.nn.functional.cross_entropy function.

loss = torch.nn.functional.cross_entropy(predicted_probabilities, true_labels)
print(loss)

Output example

tensor(0.4170)

Code explanation

1. import torch - imports the PyTorch library
2. true_labels = torch.tensor([1, 0, 0, 1]) - defines the true labels as a tensor
3. predicted_probabilities = torch.tensor([0.9, 0.2, 0.1, 0.8]) - defines the predicted probabilities as a tensor
4. loss = torch.nn.functional.cross_entropy(predicted_probabilities, true_labels) - calculates the cross entropy loss
5. print(loss) - prints out the cross entropy loss

Helpful links

• PyTorch Cross Entropy Loss
• Cross Entropy Loss Explained

Edit this code on GitHub