Broadcasting rules in TensorFlow - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

TensorFlow Broadcasting Output Shape

Given the following TensorFlow code, what is the shape of the resulting tensor after the addition?

TensorFlow

import tensorflow as tf

x = tf.constant([[1, 2, 3], [4, 5, 6]])  # shape (2, 3)
y = tf.constant([10, 20, 30])             # shape (3,)
result = x + y
result_shape = result.shape

A (2,)

B (3, 2)

C (2, 3)

D (3,)

💡 Hint

Remember how TensorFlow broadcasts smaller tensors to match larger ones by adding dimensions.

The tensor y with shape (3,) is broadcasted to (1, 3) and then to (2, 3) to match x. So the result shape is (2, 3).

🧠 Conceptual

intermediate

Understanding Broadcasting Compatibility

Which pair of tensor shapes can be broadcast together in TensorFlow without error?

A (5, 4) and (5, 4, 1)

B (4, 1, 3) and (3, 1)

C (2, 3) and (3, 2)

D (1, 2, 3) and (3, 1)

💡 Hint

Broadcasting compares shapes from the rightmost dimension to the left.

Shapes (4,1,3) and (3,1) align as (4,1,3) and (1,3,1) after prepending 1. Dimensions are compatible if equal or one is 1.

❓ Hyperparameter

advanced

Broadcasting in Tensor Operations

You have a tensor A of shape (8, 1, 6) and tensor B of shape (1, 5, 6). What will be the shape of A * B after broadcasting in TensorFlow?

A (8, 5, 6)

B (8, 5)

C (1, 5, 6)

D (8, 1, 6)

💡 Hint

Broadcasting expands dimensions where size is 1 to match the other tensor.

Tensor A shape (8,1,6) and B shape (1,5,6) broadcast to (8,5,6) by expanding the 1s.

🔧 Debug

advanced

Identifying Broadcasting Error

What error will this TensorFlow code raise due to broadcasting rules?

TensorFlow

import tensorflow as tf

x = tf.constant([[1, 2], [3, 4]])  # shape (2, 2)
y = tf.constant([1, 2, 3])          # shape (3,)
result = x + y

A No error, result shape is (2, 3)

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

C TypeError: unsupported operand type(s) for +: 'Tensor' and 'Tensor'

D InvalidArgumentError: Incompatible shapes: [2,2] vs. [3]

💡 Hint

Check if the dimensions can align from right to left for broadcasting.

Shapes (2,2) and (3,) cannot be broadcast because the last dimensions 2 and 3 differ and neither is 1.

❓ Model Choice

expert

Choosing Tensor Shapes for Efficient Broadcasting

You want to add two tensors efficiently in TensorFlow with minimal memory overhead. Which pair of shapes is best for broadcasting?

A (64, 1, 1) and (1, 32, 128)

B (64, 32, 128) and (64, 32, 128)

C (64, 32, 128) and (32, 128)

D (64, 1, 128) and (1, 32, 128)

💡 Hint

Smaller dimensions with 1 allow broadcasting without copying large data.

Option A uses many 1s allowing broadcasting with minimal memory use. Option A duplicates full shape, no broadcasting benefit.