TensorFlow · Module

tf.keras

tf.keras is TensorFlow's high-level API for building and using deep-learning models. It combines reusable layers and models with workflows for data processing, training, evaluation, tuning, and deployment.

How it works

tf.keras is TensorFlow's high-level interface for constructing, training, evaluating, and exporting deep-learning models. It organizes models from reusable layers and supplies a standard workflow through compilation, fitting, prediction, and evaluation.

Syntax

tf.keras

Key points

tf.keras is TensorFlow's high-level deep-learning interface.
The API emphasizes productive model development while retaining TensorFlow integration.
Its scope includes data preparation, model development, tuning, and deployment.
Layers and models form the central composition units.
Layers transform tensors; models connect those transformations into a directed acyclic computation graph.
Models provide built-in operations for configuring, training, predicting, and evaluating neural networks.

Common mistakes

A model is not interchangeable with an individual layer: layers perform transformations, while models organize them into a computation graph.
Deployment and execution can span specialized hardware and distributed systems, so code should not assume a single local device.

Example: Build and train a Keras model

import tensorflow as tf

inputs = tf.keras.Input(shape=(4,))
x = tf.keras.layers.Dense(8, activation="relu")(inputs)
outputs = tf.keras.layers.Dense(1)(x)
model = tf.keras.Model(inputs, outputs)
model.compile(optimizer="adam", loss="mse")
model.fit(tf.random.normal((8, 4)), tf.random.normal((8, 1)), epochs=1, verbose=0)

Builds a small functional model, configures its training behavior, and performs one training epoch.

When to use it

Model construction

Compose neural networks from reusable tensor transformations.

Training workflows

Configure and run standard training and evaluation loops.

Data processing

Prepare inputs and integrate preprocessing into models.

Deployment

Export models for scalable or cross-platform inference.

Interview questions

QWhat is tf.keras, and what problem does it solve?

tf.keras is TensorFlow's high-level API for deep learning. It provides abstractions for layers and models, plus a workflow covering model construction, training, evaluation, and deployment.

Follow-up: How does a model differ from a layer? A layer performs an input/output transformation, whereas a model represents a graph or grouping of layers and exposes training and inference operations.

QWhat are the core abstractions in tf.keras?

The main building blocks are layers and models. Layers encapsulate computation and state, while models compose layers into a directed acyclic graph and provide methods such as compile, fit, predict, and evaluate.

Follow-up: Why use the model-level API? It centralizes training configuration and execution while preserving access to TensorFlow's scaling and deployment capabilities.

QDescribe a standard tf.keras workflow.

A typical workflow creates a model from layers, configures it with compile, trains it with fit, and uses predict or evaluate afterward. Keras can also cover data processing, hyperparameter tuning, and deployment.

Follow-up: How can a Keras model be deployed beyond Python? It can be exported for browser or mobile execution and served through a web API.

tf.keras.layers.Layer
tf.keras.Model
tf.keras.Model.compile
tf.keras.Model.fit
tf.keras.Model.predict
tf.keras.Model.evaluate

Source: Keras: The high-level API for TensorFlow | TensorFlow Core. Checked 2026-10-15.