Joint and marginal distribution

For drawing n samples from joint distribution over CIFAR10 dataset, I am using:

for i, data in enumerate(trainloader, 0): 
      inputs, labels = data

    index = np.random.choice(range(inputs.shape[0]), size=batch_size, replace=False)
    outputs = inputs[index]

How can I draw n samples from marginal distribution over that? what can be the marginal distribution?