Tutorial 4

Question 1

What is torch.device() used for?

Answer 1

A torch.device is an object representing the device on which a torch.Tensor is or will be allocated.

Question 2

What is torch.cuda.is_available() used for?

Answer 2

It is used to check if a GPU is present. Returns True if so.

Question 3

What class do you have to inherit from when creating a custom neural network?

Answer 3

nn.Module()

Question 4

Which functions do you have to override?

Answer 4

def __init__() and def __forward()

Question 5

What is def __init__() for?

Answer 5

This function is used to define all of your layers.

Question 6

What is def __forward__() for?

Answer 6

This function is used to define the forward pass of your neural net.

Question 7

What does model.children() return?

Answer 7

Returns all the submodules of the model in the form of a generator

Question 8

What can you do with model.apply(f(n))

Answer 8

Apply a specific function to all the children of the model. An example is resetting the weights.

Question 9

How do you reset all the weights of a model?

Answer 9

You create a function called weight_reset(m):, which does an m.reset_parameters() call. Then you feed this function to model.apply()