Tensor shapes and reshaping in TensorFlow - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

What is the shape of the tensor after reshaping?

Given the following TensorFlow code, what is the shape of the tensor reshaped?

TensorFlow

import tensorflow as tf
original = tf.constant([[1, 2, 3], [4, 5, 6]])
reshaped = tf.reshape(original, [3, 2])
result_shape = reshaped.shape

A (3, 2)

B (2, 3)

C (6,)

D (1, 6)

💡 Hint

Remember that reshaping changes the shape but keeps the total number of elements the same.

The original tensor has shape (2, 3) with 6 elements. Reshaping to (3, 2) keeps the total elements 6 but changes the shape accordingly.

❓ Model Choice

intermediate

Choosing the correct reshape for a batch of images

You have a batch of 10 grayscale images, each 28x28 pixels, stored in a tensor of shape (10, 28, 28). Which reshape will convert this batch into a 2D tensor suitable for feeding into a fully connected layer?

A tf.reshape(tensor, [10, 784])

B tf.reshape(tensor, [10, 28, 28, 1])

C tf.reshape(tensor, [280, 28])

D tf.reshape(tensor, [28, 280])

💡 Hint

Fully connected layers expect 2D inputs: (batch_size, features).

Each 28x28 image has 784 pixels. Reshaping to (10, 784) flattens each image while keeping the batch size.

❓ Metrics

advanced

Calculating output shape after Conv2D layer

A Conv2D layer has input shape (batch_size, 64, 64, 3), kernel size (3, 3), stride 2, and 'valid' padding. What is the output shape (excluding batch size)?

A (33, 33, filters)

B (32, 32, filters)

C (31, 31, filters)

D (30, 30, filters)

💡 Hint

Output size = floor((input_size - kernel_size)/stride) + 1 for 'valid' padding.

For each spatial dimension: floor((64 - 3)/2) + 1 = floor(61/2) + 1 = 30 + 1 = 31.

🔧 Debug

advanced

Why does this reshape cause an error?

Consider this code snippet:

import tensorflow as tf
x = tf.constant([[1, 2, 3], [4, 5, 6]])
y = tf.reshape(x, [4, 2])

Why does this raise an error?

A The reshape function requires the new shape to have the same number of dimensions as the original.

B TensorFlow does not allow reshaping 2D tensors.

C The new shape dimensions must be equal to the original shape dimensions.

D The total number of elements does not match between original and new shape.

💡 Hint

Check how many elements are in the original tensor and the new shape.

Original tensor has 6 elements (2*3). New shape requires 8 elements (4*2). This mismatch causes an error.

🧠 Conceptual

expert

Effect of -1 in TensorFlow reshape

What does using -1 in a TensorFlow reshape shape argument do?

A It removes that dimension from the tensor.

B It automatically calculates the dimension size to keep the total number of elements constant.

C It sets that dimension size to zero.

D It causes a runtime error because -1 is invalid.

💡 Hint

Think about how to reshape without manually calculating one dimension.

Using -1 lets TensorFlow infer the correct size for that dimension to keep the total elements unchanged.