Estimating a sample

In previous exercises, you used linear layers to build networks.

Recall that the operation performed by nn.Linear() is to take an input \(X\) and apply the transformation \(W*X + b\) ,where \(W\) and \(b\) are two tensors (called the weight and bias).

A critical part of training PyTorch models is to calculate gradients of the weight and bias tensors with respect to a loss function.

In this exercise, you will calculate weight and bias tensor gradients using cross entropy loss and a sample of data.

The following tensors are provded:

weight: a \(2 \times 9\)-element tensor
bias: a \(2\)-element tensor
preds: a \(1 \times 2\)-element tensor containing the model predictions
target: a \(1 \times 2\)-element one-hot encoded tensor containing the ground-truth label

Use the criterion you have defined to calculate the loss value with respect to the predictions and target values.
Compute the gradients of the cross entropy loss.
Display the gradients of the weight and bias tensors, in that order.

Exercise

Estimating a sample

Instructions