Flatten and Dense layers in TensorFlow - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output shape after Flatten layer

Given the following TensorFlow model snippet, what is the shape of the tensor after the Flatten layer?

TensorFlow

import tensorflow as tf

model = tf.keras.Sequential([
    tf.keras.layers.InputLayer(input_shape=(28, 28, 3)),
    tf.keras.layers.Flatten(),
])

output_shape = model.output_shape

A (None, 3)

B (None, 28, 28, 3)

C (None, 784)

D (None, 2352)

💡 Hint

Multiply the dimensions except the batch size: 28 * 28 * 3.

The Flatten layer converts the 3D input (28x28x3) into a 1D vector per sample. 28*28*3 = 2352, so the output shape is (None, 2352).

❓ Model Choice

intermediate

Choosing the correct Dense layer output size

You want to build a classification model with 10 classes. Which Dense layer configuration correctly outputs the class scores?

A tf.keras.layers.Dense(1, activation='sigmoid')

B tf.keras.layers.Dense(10, activation='softmax')

C tf.keras.layers.Dense(10, activation='relu')

D tf.keras.layers.Dense(5, activation='softmax')

💡 Hint

For multi-class classification, output layer size equals number of classes with softmax activation.

A Dense layer with 10 units and softmax activation outputs probabilities for each of the 10 classes, suitable for multi-class classification.

❓ Hyperparameter

advanced

Effect of changing units in Dense layer

What is the effect of increasing the number of units in a Dense layer from 64 to 256 in a neural network?

A Increases model capacity and may improve learning but risks overfitting

B Has no effect on model capacity or performance

C Decreases model capacity and reduces overfitting risk

D Always guarantees better accuracy on test data

💡 Hint

More units mean more parameters and complexity.

Increasing units increases the number of parameters, allowing the model to learn more complex patterns but also increasing overfitting risk.

❓ Metrics

advanced

Interpreting accuracy with Dense output layer

A model with a Dense output layer of 3 units and softmax activation achieves 85% accuracy on validation data. What does this accuracy represent?

A The percentage of samples predicted as class 1

B The average probability output by the model

C The percentage of samples correctly classified into one of the 3 classes

D The loss value of the model on validation data

💡 Hint

Accuracy measures correct predictions over total samples.

Accuracy is the fraction of samples where the predicted class matches the true class, here among 3 classes.

🔧 Debug

expert

Identifying error in model with Flatten and Dense layers

Consider this model code snippet. What error will it raise when running model.fit()?

TensorFlow

import tensorflow as tf

model = tf.keras.Sequential([
    tf.keras.layers.Dense(64, activation='relu', input_shape=(28, 28, 3)),
    tf.keras.layers.Flatten(),
    tf.keras.layers.Dense(10, activation='softmax')
])

model.compile(optimizer='adam', loss='sparse_categorical_crossentropy')

A ValueError: Input to Dense layer must be 2D, but received 3D input

B No error, model compiles and fits correctly

C TypeError: Flatten layer cannot follow Dense layer

D RuntimeError: Output shape mismatch in last Dense layer

💡 Hint

Dense layers expect 2D inputs (batch_size, features).

The first Dense layer receives a 3D input (28,28,3) which causes a ValueError because Dense expects 2D input. Flatten should come before Dense.