Why tensors are PyTorch's core data structure - Challenge Your Understanding

Challenge - 5 Problems

🧠 Conceptual

intermediate

Why are tensors important in PyTorch?

Which of the following best explains why tensors are the core data structure in PyTorch?

A Tensors are simple Python lists that PyTorch uses to store data without any special features.

B Tensors allow PyTorch to perform fast mathematical operations on GPUs and CPUs with automatic differentiation support.

C Tensors are used only for visualization purposes in PyTorch.

D Tensors are only used to store images and cannot be used for other data types.

💡 Hint

Think about what makes PyTorch fast and flexible for machine learning.

Tensors in PyTorch are multi-dimensional arrays that support fast computation on CPUs and GPUs. They also enable automatic differentiation, which is essential for training neural networks.

❓ Predict Output

intermediate

Output of tensor operation with GPU

What will be the output of this PyTorch code snippet?

PyTorch

import torch
x = torch.tensor([1.0, 2.0, 3.0])
if torch.cuda.is_available():
    x = x.to('cuda')
print(x.device)

A Error: device not found

B cpu

C cuda:1

D cuda:0

💡 Hint

Check if CUDA is available and where the tensor is moved.

If CUDA is available, the tensor is moved to the first GPU device 'cuda:0'. Otherwise, it stays on CPU.

❓ Model Choice

advanced

Choosing tensor shape for image data

You want to input a batch of 32 RGB images of size 64x64 into a PyTorch model. Which tensor shape correctly represents this batch?

A [32, 3, 64, 64]

B [3, 32, 64, 64]

C [64, 64, 3, 32]

D [32, 64, 64, 3]

💡 Hint

PyTorch expects batch size first, then channels, then height and width.

PyTorch uses NCHW format: batch size (N), channels (C), height (H), width (W). So the correct shape is [32, 3, 64, 64].

❓ Hyperparameter

advanced

Effect of tensor dtype on model training

How does changing a tensor's data type (dtype) from float32 to float64 affect PyTorch model training?

A It has no effect on training speed or memory.

B It speeds up training and reduces memory usage.

C It increases precision but slows down training and uses more memory.

D It causes the model to ignore gradients during backpropagation.

💡 Hint

Consider the trade-off between precision and computational resources.

Float64 offers higher precision but requires more memory and computation, slowing down training.

🔧 Debug

expert

Identifying error in tensor operation

What error will this PyTorch code raise?

PyTorch

import torch
x = torch.tensor([1, 2, 3])
y = torch.tensor([4, 5])
z = x + y

A RuntimeError: The size of tensor a (3) must match the size of tensor b (2) at non-singleton dimension 0

B TypeError: unsupported operand type(s) for +: 'int' and 'list'

C ValueError: operands could not be broadcast together with shapes (3,) (2,)

D No error, output tensor is [5, 7, 3]

💡 Hint

Check if the tensors have the same shape for element-wise addition.

PyTorch requires tensors to have compatible shapes for element-wise operations. Here, shapes (3,) and (2,) mismatch causing a RuntimeError.