Keras is the high-level API of the TensorFlow platform. It provides an approachable, highly-productive interface for solving machine learning (ML) problems, with a focus on modern deep learning. Keras covers every step of the machine learning workflow, from data processing to hyperparameter tuning to deployment. It was developed with a focus on enabling fast experimentation.
With Keras, you have full access to the scalability and cross-platform capabilities of TensorFlow. You can run Keras on a TPU Pod or large clusters of GPUs, and you can export Keras models to run in the browser or on mobile devices. You can also serve Keras models via a web API.
Keras is designed to reduce cognitive load by achieving the following goals:
- Offer simple, consistent interfaces.
- Minimize the number of actions required for common use cases.
- Provide clear, actionable error messages.
- Follow the principle of progressive disclosure of complexity: It's easy to get started, and you can complete advanced workflows by learning as you go.
- Help you write concise, readable code.
Who should use Keras
The short answer is that every TensorFlow user should use the Keras APIs by default. Whether you're an engineer, a researcher, or an ML practitioner, you should start with Keras.
There are a few use cases (for example, building tools on top of TensorFlow or developing your own high-performance platform) that require the low-level TensorFlow Core APIs. But if your use case doesn't fall into one of the Core API applications, you should prefer Keras.
Keras API components
tf.keras.layers.Layer class is the fundamental abstraction in Keras. A
Layer encapsulates a state (weights) and some computation (defined in the
Weights created by layers can be trainable or non-trainable. Layers are recursively composable: If you assign a layer instance as an attribute of another layer, the outer layer will start tracking the weights created by the inner layer.
You can also use layers to handle data preprocessing tasks like normalization and text vectorization. Preprocessing layers can be included directly into a model, either during or after training, which makes the model portable.
A model is an object that groups layers together and that can be trained on data.
The simplest type of model is the
which is a linear stack of layers. For more complex architectures, you can
either use the
Keras functional API,
which lets you build arbitrary graphs of layers, or
use subclassing to write models from scratch.
tf.keras.Model class features built-in training and evaluation methods:
tf.keras.Model.fit: Trains the model for a fixed number of epochs.
tf.keras.Model.predict: Generates output predictions for the input samples.
tf.keras.Model.evaluate: Returns the loss and metrics values for the model; configured via the
These methods give you access to the following built-in training features:
- Callbacks. You can leverage built-in callbacks for early stopping, model checkpointing, and TensorBoard monitoring. You can also implement custom callbacks.
- Distributed training. You can easily scale up your training to multiple GPUs, TPUs, or devices.
- Step fusing. With the
tf.keras.Model.compile, you can process multiple batches in a single
tf.functioncall, which greatly improves device utilization on TPUs.
Other APIs and tools
Keras provides many other APIs and tools for deep learning, including:
To get started using Keras with TensorFlow, check out the following topics:
- The Sequential model
- The Functional API
- Training & evaluation with the built-in methods
- Making new layers and models via subclassing
- Serialization and saving
- Working with preprocessing layers
- Customizing what happens in fit()
- Writing a training loop from scratch
- Working with RNNs
- Understanding masking & padding
- Writing your own callbacks
- Transfer learning & fine-tuning
- Multi-GPU and distributed training
To learn more about Keras, see the following topics at keras.io: