How can TensorFlow be used to define a loss function, an optimizer, train the model and evaluate it on the IMDB dataset in Python?

TensorFlow is a machine learning framework provided by Google. It is an open-source framework used in conjunction with Python to implement algorithms, deep learning applications and much more. It is used in research and for production purposes.

The IMDB dataset contains reviews of over 50,000 movies and is commonly used for Natural Language Processing tasks, particularly sentiment analysis. This tutorial demonstrates how to define a loss function, optimizer, train a model, and evaluate it using TensorFlow.

Installation

The 'tensorflow' package can be installed using the following command ?

pip install tensorflow

Model Configuration

Before training, the model needs to be compiled with a loss function, optimizer, and metrics. Here's how to configure these components ?

import tensorflow as tf
from tensorflow.keras import losses

# Compile the model with loss function, optimizer, and metrics
model.compile(
    loss=losses.BinaryCrossentropy(from_logits=True),
    optimizer='adam',
    metrics=tf.metrics.BinaryAccuracy(threshold=0.0)
)

Components Explained

• Loss Function: BinaryCrossentropy is used for binary classification tasks like sentiment analysis

• Optimizer: adam is an efficient gradient descent algorithm

• Metrics: BinaryAccuracy measures the percentage of predictions that match the binary labels

Training the Model

Once compiled, train the model using the training and validation datasets ?

# Set number of training epochs
epochs = 10

# Train the model
history = model.fit(
    train_ds,
    validation_data=val_ds,
    epochs=epochs
)

Model Evaluation

After training, evaluate the model's performance on the test dataset ?

# Evaluate model on test data
loss, accuracy = model.evaluate(test_ds)

print("Loss is :", loss)
print("Accuracy is :", accuracy)

Training Output

The training process shows progress for each epoch with loss and accuracy metrics ?

Epoch 1/10
625/625 [==============================] - 12s 19ms/step - loss: 0.6818 - binary_accuracy: 0.6130 - val_loss: 0.6135 - val_binary_accuracy: 0.7750
Epoch 2/10
625/625 [==============================] - 4s 7ms/step - loss: 0.5785 - binary_accuracy: 0.7853 - val_loss: 0.4971 - val_binary_accuracy: 0.8230
Epoch 3/10
625/625 [==============================] - 4s 7ms/step - loss: 0.4651 - binary_accuracy: 0.8372 - val_loss: 0.4193 - val_binary_accuracy: 0.8470
Epoch 4/10
625/625 [==============================] - 4s 7ms/step - loss: 0.3901 - binary_accuracy: 0.8635 - val_loss: 0.3732 - val_binary_accuracy: 0.8612
Epoch 5/10
625/625 [==============================] - 4s 7ms/step - loss: 0.3435 - binary_accuracy: 0.8771 - val_loss: 0.3444 - val_binary_accuracy: 0.8688
Epoch 6/10
625/625 [==============================] - 4s 7ms/step - loss: 0.3106 - binary_accuracy: 0.8877 - val_loss: 0.3255 - val_binary_accuracy: 0.8730
Epoch 7/10
625/625 [==============================] - 5s 7ms/step - loss: 0.2855 - binary_accuracy: 0.8970 - val_loss: 0.3119 - val_binary_accuracy: 0.8732
Epoch 8/10
625/625 [==============================] - 5s 7ms/step - loss: 0.2652 - binary_accuracy: 0.9048 - val_loss: 0.3027 - val_binary_accuracy: 0.8772
Epoch 9/10
625/625 [==============================] - 5s 7ms/step - loss: 0.2481 - binary_accuracy: 0.9125 - val_loss: 0.2959 - val_binary_accuracy: 0.8782
Epoch 10/10
625/625 [==============================] - 5s 7ms/step - loss: 0.2328 - binary_accuracy: 0.9161 - val_loss: 0.2913 - val_binary_accuracy: 0.8792
782/782 [==============================] - 10s 12ms/step - loss: 0.3099 - binary_accuracy: 0.8741
Loss is : 0.3099007308483124
Accuracy is : 0.8741199970245361

Key Observations

• Training loss decreases from 0.68 to 0.23 over 10 epochs

• Training accuracy improves from 61% to 91%

• Final test accuracy reaches 87.4%

• Validation metrics help monitor overfitting during training

Conclusion

TensorFlow's compile(), fit(), and evaluate() methods provide a complete workflow for training neural networks. The model achieved 87.4% accuracy on the IMDB sentiment classification task, demonstrating effective learning from the training data.