Input shape specification in TensorFlow - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

What is the output shape of the model's first layer?

Consider the following TensorFlow Keras model code. What is the output shape of the first Dense layer?

TensorFlow

import tensorflow as tf
model = tf.keras.Sequential([
    tf.keras.layers.InputLayer(input_shape=(28, 28)),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(64, activation='relu')
])
output_shape = model.layers[2].output_shape

A (None, 28, 28, 64)

B (None, 64)

C (64,)

D (28, 28, 64)

💡 Hint

Remember that Flatten changes the input shape before the Dense layer.

The Flatten layer converts the input from (28, 28) to (784,), so the Dense layer outputs (None, 64) where None is the batch size.

❓ Model Choice

intermediate

Which input shape is correct for grayscale 32x32 images?

You want to build a CNN model in TensorFlow for grayscale images of size 32x32 pixels. Which input_shape argument is correct for the first Conv2D layer?

A (32, 32)

B (1, 32, 32)

C (32, 32, 1)

D (None, 32, 32, 1)

💡 Hint

TensorFlow expects channels last format by default.

TensorFlow uses channels last format, so the input shape should be (height, width, channels). For grayscale, channels=1.

❓ Hyperparameter

advanced

What happens if input_shape is omitted in the first layer?

In TensorFlow Keras, what is the effect of not specifying input_shape in the first layer of a Sequential model?

A Model will automatically add an InputLayer with shape (None, None).

B Model will raise an error when building because input shape is unknown.

C Model will use a default input shape of (None, 1).

D Model will infer input shape from the first batch of data during training.

💡 Hint

Think about how Keras can build models dynamically.

If input_shape is not specified, Keras waits until it sees data to infer the input shape automatically.

🔧 Debug

advanced

Why does this model raise a shape error?

Given this model code, why does it raise a shape mismatch error when training?

TensorFlow

import tensorflow as tf
model = tf.keras.Sequential([
    tf.keras.layers.Dense(10, activation='relu', input_shape=(20,)),
    tf.keras.layers.Dense(5)
])

import numpy as np
x = np.random.rand(32, 10)
y = np.random.rand(32, 5)
model.compile(optimizer='adam', loss='mse')
model.fit(x, y, epochs=1)

A Input data x has shape (32, 10) but model expects input shape (20,), causing mismatch.

B Output data y has wrong shape; it should be (32, 10) to match last layer.

C The Dense layers require 3D input but x is 2D.

D The model is missing an activation function in the last layer.

💡 Hint

Check the input_shape argument and the shape of x.

The model expects input vectors of length 20, but x has vectors of length 10, causing a shape mismatch error.

🧠 Conceptual

expert

Why specify batch size in Input layer for stateful RNNs?

In TensorFlow, why is it necessary to specify batch_size in the Input layer when building a stateful RNN model?

A Because stateful RNNs keep states between batches, fixed batch size is needed to maintain state consistency.

B Because batch_size controls the number of neurons in the RNN layer.

C Because batch_size defines the sequence length for the RNN input.

D Because batch_size is required to enable GPU acceleration.

💡 Hint

Think about how RNN states are preserved across batches.

Stateful RNNs keep hidden states between batches, so batch size must be fixed and known to maintain correct state alignment.