Multi-input and multi-output models in TensorFlow - Model Metrics & Evaluation

Metrics & Evaluation - Multi-input and multi-output models

Which metric matters for Multi-input and Multi-output models and WHY

When a model has multiple inputs and outputs, each output can represent a different task. So, we need to measure performance separately for each output. For example, if one output predicts a category (classification), accuracy or F1-score matters. If another output predicts a number (regression), mean squared error (MSE) matters. This helps us understand how well the model does on each task.

Confusion matrix or equivalent visualization

For classification outputs, a confusion matrix shows how many predictions were correct or wrong for each class. For example, if one output predicts "cat" or "dog":

      | Predicted Cat | Predicted Dog |
      |--------------|---------------|
      | True Cat: 50 | False Dog: 3  |
      | False Cat: 5 | True Dog: 42  |
    

This helps calculate precision, recall, and F1 for that output. For regression outputs, we look at error values like MSE instead.

Precision vs Recall tradeoff with concrete examples

Imagine a multi-output model where one output detects spam emails (classification) and another predicts email length (regression). For spam detection, high precision means few good emails are wrongly marked as spam. High recall means most spam emails are caught. Depending on what matters more, we adjust the model or threshold.

For the length prediction output, we focus on minimizing error, not precision or recall.

What "good" vs "bad" metric values look like for this use case

Good metrics mean each output performs well on its task. For classification outputs, precision and recall above 0.8 are usually good. For regression outputs, low MSE or MAE (mean absolute error) is good.

Bad metrics are low precision or recall (below 0.5) for classification, or very high error for regression. This means the model struggles on that output.

Common pitfalls in metrics for multi-input and multi-output models

• Ignoring some outputs: Only checking metrics for one output hides problems in others.
• Mixing metric types: Using accuracy for regression outputs or MSE for classification is wrong.
• Data leakage: If inputs share information that leaks target info, metrics look too good.
• Overfitting: High training metrics but poor validation metrics on any output means overfitting.

Self-check question

Your multi-output model has 98% accuracy on one classification output but only 12% recall on detecting fraud in another output. Is it good for production? Why or why not?

Answer: No, it is not good. Even though accuracy is high on one output, the very low recall on fraud detection means the model misses most fraud cases. For fraud detection, recall is critical because missing fraud is costly. So, the model needs improvement on that output before production.

Key Result

Evaluate each output separately using the right metric type; high recall is critical for sensitive tasks like fraud detection.

Practice

1. What is the main purpose of a multi-input model in TensorFlow?

easy

A. To accept more than one data source at the same time

B. To predict multiple outputs from a single input

C. To train faster using GPU acceleration

D. To reduce the number of layers in the model

Solution

1. Step 1: Understand multi-input models

   Multi-input models are designed to take multiple data sources as inputs simultaneously.

2. Step 2: Differentiate from multi-output models

   Multi-output models predict multiple outputs but usually from a single input source.

3. Final Answer:

   To accept more than one data source at the same time -> Option A

4. Quick Check:

   Multi-input = multiple data sources [OK]

Hint: Multi-input means many inputs, not many outputs [OK]

Common Mistakes:

• Confusing multi-input with multi-output
• Thinking multi-input reduces layers
• Assuming multi-input speeds training automatically

2. Which of the following is the correct way to define two inputs in a TensorFlow Keras model?

easy

A. inputs = tf.keras.Input(shape=(10, 5))

B. inputs = tf.keras.Input(shape=(10,)), tf.keras.Input(shape=(5,))

C. inputs = [tf.keras.Input(shape=(10,)), tf.keras.Input(shape=(5,))]

D. inputs = tf.keras.Input(shape=(10,)); inputs = tf.keras.Input(shape=(5,))

Solution

1. Step 1: Recall how to define multiple inputs

   Multiple inputs should be stored as a list of Input layers in Keras.

2. Step 2: Check each option

   inputs = [tf.keras.Input(shape=(10,)), tf.keras.Input(shape=(5,))] correctly creates a list of two Input layers. inputs = tf.keras.Input(shape=(10,)), tf.keras.Input(shape=(5,)) creates a tuple but does not assign it properly. inputs = tf.keras.Input(shape=(10, 5)) defines a single input with combined shape. inputs = tf.keras.Input(shape=(10,)); inputs = tf.keras.Input(shape=(5,)) overwrites the first input with the second.

3. Final Answer:

   inputs = [tf.keras.Input(shape=(10,)), tf.keras.Input(shape=(5,))] -> Option C

4. Quick Check:

   Multiple inputs = list of Input layers [OK]

Hint: Use a list to hold multiple Input layers [OK]

Common Mistakes:

• Using a tuple instead of a list for inputs
• Overwriting inputs instead of storing both
• Combining shapes into one input incorrectly

3. What will be the output shape of the following multi-output model?

input1 = tf.keras.Input(shape=(8,))
input2 = tf.keras.Input(shape=(4,))
x1 = tf.keras.layers.Dense(5)(input1)
x2 = tf.keras.layers.Dense(3)(input2)
output1 = tf.keras.layers.Dense(2)(x1)
output2 = tf.keras.layers.Dense(1)(x2)
model = tf.keras.Model(inputs=[input1, input2], outputs=[output1, output2])
print([o.shape for o in model.outputs])

medium

A. [TensorShape([None, 2]), TensorShape([None, 1])]

B. [TensorShape([8, 2]), TensorShape([4, 1])]

C. [TensorShape([None, 5]), TensorShape([None, 3])]

D. [TensorShape([None, 8]), TensorShape([None, 4])]

Solution

1. Step 1: Trace output layers shapes

   output1 is Dense(2) applied to x1, so shape is (None, 2). output2 is Dense(1) applied to x2, so shape is (None, 1).

2. Step 2: Understand batch dimension

   TensorFlow uses None for batch size, so output shapes include None as first dimension.

3. Final Answer:

   [TensorShape([None, 2]), TensorShape([None, 1])] -> Option A

4. Quick Check:

   Output shapes match Dense layer units [OK]

Hint: Output shape = batch size null + Dense units [OK]

Common Mistakes:

• Confusing input shape with output shape
• Ignoring batch dimension null
• Mixing intermediate layer shapes with output shapes

4. Identify the error in this multi-output model definition:

input = tf.keras.Input(shape=(10,))
x = tf.keras.layers.Dense(8)(input)
output1 = tf.keras.layers.Dense(4)(x)
output2 = tf.keras.layers.Dense(3)(x)
model = tf.keras.Model(inputs=input, outputs=[output1, output2])

medium

A. inputs should be a list when there are multiple outputs

B. outputs should be a single tensor, not a list

C. inputs must be a list even if only one input exists

D. No error, the model is defined correctly

Solution

1. Step 1: Check inputs parameter

   inputs can be a single Input layer if there is only one input source.

2. Step 2: Check outputs parameter

   outputs can be a list of tensors to define multiple outputs.

3. Final Answer:

   No error, the model is defined correctly -> Option D

4. Quick Check:

   Single input + multiple outputs = valid model [OK]

Hint: Single input can be passed directly, outputs can be list [OK]

Common Mistakes:

• Thinking inputs must always be a list
• Believing outputs cannot be a list
• Assuming multiple outputs require multiple inputs

5. You want to build a model that takes two inputs: an image (shape 64x64x3) and a vector of 10 features. It should output two predictions: a 5-class classification and a single continuous value. Which is the correct way to define the model inputs and outputs?

hard

A. inputs = [tf.keras.Input(shape=(64,64,3)), tf.keras.Input(shape=(10,))]; outputs = tf.keras.layers.Dense(6)(inputs)

B. inputs = [tf.keras.Input(shape=(64,64,3)), tf.keras.Input(shape=(10,))]; outputs = [tf.keras.layers.Dense(5, activation='softmax')(x), tf.keras.layers.Dense(1)(y)]

C. inputs = tf.keras.Input(shape=(64,64,3)); outputs = [tf.keras.layers.Dense(5)(inputs), tf.keras.layers.Dense(1)(inputs)]

D. inputs = tf.keras.Input(shape=(74,)); outputs = [tf.keras.layers.Dense(5)(inputs), tf.keras.layers.Dense(1)(inputs)]

Solution

1. Step 1: Define inputs separately for image and vector

   Two inputs require two Input layers with correct shapes: (64,64,3) for image and (10,) for vector.

2. Step 2: Define outputs separately for classification and regression

   Outputs are two layers: one Dense with 5 units and softmax for classification, one Dense with 1 unit for continuous value.

3. Final Answer:

   inputs = [tf.keras.Input(shape=(64,64,3)), tf.keras.Input(shape=(10,))]; outputs = [tf.keras.layers.Dense(5, activation='softmax')(x), tf.keras.layers.Dense(1)(y)] -> Option B

4. Quick Check:

   Separate inputs and outputs for multi-input/output model [OK]

Hint: Match each input and output with separate Input and Dense layers [OK]

Common Mistakes:

• Combining inputs into one vector incorrectly
• Using single input for different data types
• Outputting combined units instead of separate outputs