Chapter 13: TensorFlow Keras API – Anatomy of a Model

“A model is just an idea—until it gets layers, weights, and shape.”

---

13.1 What is tf.keras?

tf.keras is TensorFlow’s official high-level API for building, training, and deploying machine learning models.

It's designed to be:

• User-friendly (simple syntax)
• Modular (layers, optimizers, callbacks)
• Extensible (custom layers/models)
• Integrated (with TensorFlow ecosystem)

Keras wraps the complexity of TensorFlow so you can focus on structure and logic, not boilerplate.

---

13.2 The 3 Model Building Styles

There are three ways to build models using tf.keras:

✅ 1. Sequential API (Beginner-friendly)

from tensorflow.keras import layers, models

model = models.Sequential([
    layers.Dense(64, activation='relu', input_shape=(100,)),
    layers.Dense(10, activation='softmax')
])

✅ 2. Functional API (Flexible architectures)

inputs = tf.keras.Input(shape=(100,))
x = layers.Dense(64, activation='relu')(inputs)
outputs = layers.Dense(10, activation='softmax')(x)

model = tf.keras.Model(inputs=inputs, outputs=outputs)

✅ 3. Subclassing API (For full control)

class MyModel(tf.keras.Model):
    def __init__(self):
        super().__init__()
        self.dense1 = layers.Dense(64, activation='relu')
        self.out = layers.Dense(10, activation='softmax')

    def call(self, x):
        x = self.dense1(x)
        return self.out(x)

model = MyModel()

Each has trade-offs. Start with Sequential, move to Functional for branching inputs/outputs, and use Subclassing for full customization.

---

13.3 Anatomy of a Keras Model

Here’s what makes up a model under the hood:

Component	Description
Input Layer	Defines the shape of input data
Hidden Layers	The intermediate processing units
Output Layer	Final layer for predictions
Loss Function	Measures model’s error
Optimizer	Updates weights based on gradients
Metrics	Monitors performance (accuracy, loss, etc.)

---

13.4 Model Compilation

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

This sets the training configuration, including how the model learns and what it tracks.

---

13.5 Summary of a Simple Model Lifecycle

# 1. Build the model
model = tf.keras.Sequential([
    layers.Dense(128, activation='relu'),
    layers.Dense(10, activation='softmax')
])

# 2. Compile
model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['accuracy'])

# 3. Train
model.fit(x_train, y_train, epochs=5)

# 4. Evaluate
model.evaluate(x_test, y_test)

# 5. Predict
preds = model.predict(x_new)

---

13.6 Summary

• tf.keras is TensorFlow’s high-level API for model building.
• You can build models using Sequential, Functional, or Subclassing styles.
• Models have layers, losses, optimizers, and metrics—all handled cleanly.
• Knowing the anatomy helps you debug, customize, and scale efficiently.

---

“A model is just an idea—until it gets layers, weights, and shape.”