Multi-input and multi-output models in TensorFlow - Practice Problems & Coding Challenges

Challenge - 5 Problems

❓ Predict Output

intermediate

Output shape of a multi-input model

Consider a TensorFlow model with two inputs: one of shape (None, 10) and another of shape (None, 5). Both inputs are concatenated and passed through a Dense layer with 8 units. What is the shape of the model's output?

TensorFlow

import tensorflow as tf

input1 = tf.keras.Input(shape=(10,))
input2 = tf.keras.Input(shape=(5,))

concat = tf.keras.layers.Concatenate()([input1, input2])
output = tf.keras.layers.Dense(8)(concat)

model = tf.keras.Model(inputs=[input1, input2], outputs=output)

print(model.output_shape)

A (None, 8)

B (None, 10)

C (None, 15)

D (None, 5)

💡 Hint

The Dense layer outputs a tensor with shape (batch_size, units).

The two inputs are concatenated along the last axis, resulting in shape (None, 15). The Dense layer with 8 units transforms this to (None, 8).

❓ Model Choice

intermediate

Choosing the correct multi-output model architecture

You want to build a TensorFlow model that takes one input of shape (20,) and produces two outputs: one for regression (a single continuous value) and one for classification (3 classes). Which model architecture below correctly implements this?

A One input layer, two Dense layers with 3 units each and relu activation for outputs.

B Two input layers, one output layer with 4 units and softmax activation.

C One input layer, one Dense layer with 4 units and sigmoid activation for both outputs.

D One input layer, shared Dense layers, then two separate Dense layers: one with 1 unit (no activation) for regression, one with 3 units and softmax for classification.

💡 Hint

Regression output usually has 1 unit without activation; classification output uses softmax for multiple classes.

Option D correctly uses one input, shared layers, and two output heads: one for regression (1 unit, linear) and one for classification (3 units, softmax).

❓ Hyperparameter

advanced

Choosing loss functions for multi-output models

You have a multi-output model with two outputs: a regression output and a 4-class classification output. Which combination of loss functions is appropriate to compile this model?

A {'regression_output': 'mse', 'classification_output': 'categorical_crossentropy'}

B {'regression_output': 'binary_crossentropy', 'classification_output': 'mse'}

C {'regression_output': 'categorical_crossentropy', 'classification_output': 'mse'}

D {'regression_output': 'mse', 'classification_output': 'binary_crossentropy'}

💡 Hint

Regression uses mean squared error; multi-class classification uses categorical crossentropy.

Mean squared error (mse) is suitable for regression outputs. Categorical crossentropy is used for multi-class classification with one-hot labels.

🔧 Debug

advanced

Debugging multi-input model input mismatch error

You defined a TensorFlow model with two inputs: input1 shape (None, 8) and input2 shape (None, 4). When training, you get a ValueError about input shapes not matching. Which of the following is the most likely cause?

A You used different batch sizes for input1 and input2 arrays.

B You passed a single numpy array instead of a list or tuple of two arrays as input data.

C You forgot to compile the model before training.

D You used a Dense layer instead of Conv2D in the model.

💡 Hint

Multi-input models expect inputs as a list or tuple matching the input layers.

Passing a single array instead of a list/tuple of arrays causes shape mismatch errors because the model expects multiple inputs.

🧠 Conceptual

expert

Understanding training metrics in multi-output models

A multi-output model has two outputs: output A (regression) and output B (classification). You compile the model with metrics={'output_A': ['mae'], 'output_B': ['accuracy']}. During training, what will the reported 'accuracy' metric represent?

A The overall accuracy combining regression and classification outputs.

B The average accuracy of both outputs A and B.

C The classification accuracy of output B only.

D The mean absolute error of output A converted to accuracy.

💡 Hint

Metrics are reported per output when specified as a dictionary.

The 'accuracy' metric applies only to output B (classification). Output A uses 'mae' metric separately.