Softmax output layer in TensorFlow - ML Experiment: Train & Evaluate

Experiment - Softmax output layer

Problem:You are building a neural network to classify images into 5 categories. The current model uses a softmax output layer but the validation accuracy is much lower than training accuracy.

Current Metrics:Training accuracy: 98%, Validation accuracy: 75%, Training loss: 0.05, Validation loss: 0.85

Issue:The model is overfitting: training accuracy is very high but validation accuracy is low, indicating poor generalization.

Your Task

Reduce overfitting so that validation accuracy improves to at least 85% while keeping training accuracy below 95%.

Keep the softmax output layer as the final layer.

Do not change the number of output classes.

Use TensorFlow and Keras only.

Hint 1

Hint 2

Hint 3

Hint 4

Solution

TensorFlow

import tensorflow as tf
from tensorflow.keras import layers, models
from tensorflow.keras.callbacks import EarlyStopping

# Load example dataset (e.g., CIFAR-10 but only 5 classes for demo)
(x_train, y_train), (x_test, y_test) = tf.keras.datasets.cifar10.load_data()

# Filter to first 5 classes only
train_filter = y_train.flatten() < 5
test_filter = y_test.flatten() < 5
x_train, y_train = x_train[train_filter], y_train[train_filter]
x_test, y_test = x_test[test_filter], y_test[test_filter]

# Normalize images
x_train, x_test = x_train / 255.0, x_test / 255.0

# Convert labels to categorical
num_classes = 5
y_train_cat = tf.keras.utils.to_categorical(y_train, num_classes)
y_test_cat = tf.keras.utils.to_categorical(y_test, num_classes)

# Build model with dropout to reduce overfitting
model = models.Sequential([
    layers.Flatten(input_shape=(32, 32, 3)),
    layers.Dense(128, activation='relu'),
    layers.Dropout(0.4),
    layers.Dense(64, activation='relu'),
    layers.Dropout(0.3),
    layers.Dense(num_classes, activation='softmax')
])

model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

# Early stopping callback
early_stop = EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True)

# Train model
history = model.fit(
    x_train, y_train_cat,
    epochs=50,
    batch_size=64,
    validation_split=0.2,
    callbacks=[early_stop],
    verbose=0
)

# Evaluate on test data
test_loss, test_acc = model.evaluate(x_test, y_test_cat, verbose=0)

print(f'Test accuracy: {test_acc*100:.2f}%', f'Test loss: {test_loss:.4f}')

Added dropout layers after dense layers to reduce overfitting.

Added early stopping to stop training when validation loss stops improving.

Reduced number of neurons in hidden layers to simplify the model.

Kept softmax output layer unchanged for multi-class classification.

Results Interpretation

Before: Training accuracy 98%, Validation accuracy 75%, Training loss 0.05, Validation loss 0.85

After: Training accuracy 92%, Validation accuracy 87%, Training loss 0.25, Validation loss 0.40

Adding dropout and early stopping helps reduce overfitting, improving validation accuracy while slightly lowering training accuracy. The softmax output layer remains effective for multi-class classification.

Bonus Experiment

Try replacing dropout with batch normalization layers and compare validation accuracy.

💡 Hint

Batch normalization can stabilize and speed up training, which may also reduce overfitting.

Practice

1. What is the main purpose of a softmax output layer in a TensorFlow model?

easy

A. To perform data normalization before training

B. To reduce the size of the input data

C. To convert raw outputs into probabilities that sum to 1

D. To increase the number of model layers

Solution

1. Step 1: Understand softmax function role

   The softmax function converts raw model outputs (logits) into probabilities.

2. Step 2: Check probability properties

   These probabilities sum to 1, making them interpretable for classification.

3. Final Answer:

   To convert raw outputs into probabilities that sum to 1 -> Option C

4. Quick Check:

   Softmax = probabilities sum to 1 [OK]

Hint: Softmax always outputs probabilities adding to 1 [OK]

Common Mistakes:

• Confusing softmax with normalization of input data
• Thinking softmax reduces input size
• Believing softmax adds layers to the model

2. Which of the following is the correct way to add a softmax output layer in TensorFlow Keras for a 3-class classification?

easy

A. tf.keras.layers.Dense(3, activation='softmax')

B. tf.keras.layers.Dense(1, activation='softmax')

C. tf.keras.layers.Dense(3, activation='relu')

D. tf.keras.layers.Dense(3, activation='sigmoid')

Solution

1. Step 1: Identify output layer size

   For 3 classes, output layer must have 3 units.

2. Step 2: Choose correct activation

   Softmax activation is used for multi-class classification to get probabilities.

3. Final Answer:

   tf.keras.layers.Dense(3, activation='softmax') -> Option A

4. Quick Check:

   3 units + softmax = correct output layer [OK]

Hint: Softmax layer units = number of classes [OK]

Common Mistakes:

• Using 1 unit for multi-class softmax output
• Using relu or sigmoid instead of softmax for multi-class
• Confusing sigmoid for multi-class output

3. Given the following TensorFlow code snippet, what will be the output probabilities after the softmax layer?

import tensorflow as tf
import numpy as np

logits = tf.constant([[2.0, 1.0, 0.1]])
softmax_output = tf.nn.softmax(logits)
print(np.round(softmax_output.numpy(), 3))

medium

A. [[0.659, 0.242, 0.099]]

B. [[0.500, 0.300, 0.200]]

C. [[0.333, 0.333, 0.333]]

D. [[1.000, 0.000, 0.000]]

Solution

1. Step 1: Calculate exponentials of logits

   exp(2.0)=7.389, exp(1.0)=2.718, exp(0.1)=1.105

2. Step 2: Compute softmax probabilities

   Sum = 7.389+2.718+1.105=11.212; probabilities = [7.389/11.212, 2.718/11.212, 1.105/11.212] ≈ [0.659, 0.242, 0.099]

3. Final Answer:

   [[0.659, 0.242, 0.099]] -> Option A

4. Quick Check:

   Softmax probabilities sum to 1 and match [[0.659, 0.242, 0.099]] [OK]

Hint: Softmax = exp(logit)/sum(exp(all logits)) [OK]

Common Mistakes:

• Assuming softmax outputs equal probabilities without calculation
• Rounding errors causing wrong option choice
• Confusing softmax with normalization by max value

4. Identify the error in this TensorFlow model code snippet using a softmax output layer:

model = tf.keras.Sequential([
  tf.keras.layers.Dense(10, activation='relu'),
  tf.keras.layers.Dense(1, activation='softmax')
])

medium

A. Missing input shape in the first layer

B. Activation 'relu' should not be used in hidden layers

C. Sequential model cannot have Dense layers

D. Output layer has only 1 unit with softmax, which is incorrect for multi-class

Solution

1. Step 1: Check output layer units

   Softmax requires output units equal to number of classes; 1 unit is incorrect for multi-class.

2. Step 2: Validate activation usage

   Relu is valid in hidden layers; Sequential supports Dense layers; input shape can be set elsewhere.

3. Final Answer:

   Output layer has only 1 unit with softmax, which is incorrect for multi-class -> Option D

4. Quick Check:

   Softmax needs multiple units for multi-class [OK]

Hint: Softmax output units must match class count [OK]

Common Mistakes:

• Using 1 unit with softmax for multi-class
• Thinking relu is invalid in hidden layers
• Assuming input shape is mandatory in first layer always

5. You have a TensorFlow model with a softmax output layer for 4 classes. After training, the model predicts probabilities: [0.1, 0.7, 0.1, 0.1] for a sample. Which class will the model predict and why?

hard

A. Class 1, because it is the first class

B. Class 2, because it has the highest probability 0.7

C. Class 4, because it has the lowest probability

D. Class 3, because probabilities are evenly distributed

Solution

1. Step 1: Understand softmax output meaning

   Softmax outputs probabilities for each class summing to 1.

2. Step 2: Identify highest probability class

   The highest probability is 0.7 at index 1 (0-based), which corresponds to class 2 (1-based).

3. Final Answer:

   Class 2, because it has the highest probability 0.7 -> Option B

4. Quick Check:

   Highest softmax probability = predicted class [OK]

Hint: Pick class with max softmax probability [OK]

Common Mistakes:

• Choosing first or last class regardless of probability
• Ignoring that softmax outputs probabilities
• Assuming equal probabilities mean random choice