Keras
Restez organisé à l'aide des collections
Enregistrez et classez les contenus selon vos préférences.
tf.keras est l'API de haut niveau de TensorFlow permettant de créer et d'entraîner des modèles de deep learning. Elle est utilisée dans le cadre du prototypage rapide, de la recherche de pointe et du passage en production. Elle présente trois avantages majeurs :
- Convivialité
Keras dispose d'une interface simple et cohérente, optimisée pour les cas d'utilisation courants. Elle fournit des informations claires et concrètes concernant les erreurs des utilisateurs.
- Modularité et facilité de composition
Les modèles Keras sont créés en connectant des composants configurables, avec quelques restrictions.
- Facilité d'extension
Composez des éléments de base personnalisés pour exprimer de nouvelles idées de recherche. Créez des calques, des métriques et des fonctions de perte, et développez des modèles de pointe.
Le guide intitulé Keras: A Quick Overview (Présentation rapide de Keras) vous aidera à faire vos premiers pas.
Pour une présentation du machine learning avec tf.keras destinée aux utilisateurs novices, consultez cet ensemble de tutoriels de démarrage.
Pour une présentation détaillée de l'API, consultez les guides suivants qui contiennent tout ce que vous devez savoir en tant qu'utilisateur expérimenté de TensorFlow Keras :
Regardez la série de vidéos Inside TensorFlow sur YouTube pour une présentation détaillée du fonctionnement interne de Keras :
Sauf indication contraire, le contenu de cette page est régi par une licence Creative Commons Attribution 4.0, et les échantillons de code sont régis par une licence Apache 2.0. Pour en savoir plus, consultez les Règles du site Google Developers. Java est une marque déposée d'Oracle et/ou de ses sociétés affiliées.
Dernière mise à jour le 2020/06/05 (UTC).
[null,null,["Dernière mise à jour le 2020/06/05 (UTC)."],[],[],null,["# Keras: The high-level API for TensorFlow\n\n\u003cbr /\u003e\n\nKeras is the high-level API of the TensorFlow platform. It provides an\napproachable, highly-productive interface for solving machine learning (ML)\nproblems, with a focus on modern deep learning. Keras covers every step of the\nmachine learning workflow, from data processing to hyperparameter tuning to\ndeployment. It was developed with a focus on enabling fast experimentation.\n\nWith Keras, you have full access to the scalability and cross-platform\ncapabilities of TensorFlow. You can run Keras on a TPU Pod or large clusters of\nGPUs, and you can export Keras models to run in the browser or on mobile\ndevices. You can also serve Keras models via a web API.\n\nKeras is designed to reduce cognitive load by achieving the following goals:\n\n- Offer simple, consistent interfaces.\n- Minimize the number of actions required for common use cases.\n- Provide clear, actionable error messages.\n- 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.\n- Help you write concise, readable code.\n\nWho should use Keras\n--------------------\n\nThe short answer is that every TensorFlow user should use the Keras APIs by\ndefault. Whether you're an engineer, a researcher, or an ML practitioner, you\nshould start with Keras.\n\nThere are a few use cases (for example, building tools on top of TensorFlow or\ndeveloping your own high-performance platform) that require the low-level\n[TensorFlow Core APIs](https://www.tensorflow.org/guide/core). But if your use\ncase doesn't fall into one\nof the\n[Core API applications](https://www.tensorflow.org/guide/core#core_api_applications),\nyou should prefer Keras.\n\nKeras API components\n--------------------\n\nThe core data structures of Keras are [layers](https://keras.io/api/layers/) and\n[models](https://keras.io/api/models/). A layer is a simple input/output\ntransformation, and a model is a directed acyclic graph (DAG) of layers.\n\n### Layers\n\nThe `tf.keras.layers.Layer` class is the fundamental abstraction in Keras. A\n`Layer` encapsulates a state (weights) and some computation (defined in the\n`tf.keras.layers.Layer.call` method).\n\nWeights created by layers can be trainable or non-trainable. Layers are\nrecursively composable: If you assign a layer instance as an attribute of\nanother layer, the outer layer will start tracking the weights created by the\ninner layer.\n\nYou can also use layers to handle data preprocessing tasks like normalization\nand text vectorization. Preprocessing layers can be included directly into a\nmodel, either during or after training, which makes the model portable.\n\n### Models\n\nA model is an object that groups layers together and that can be trained on\ndata.\n\nThe simplest type of model is the\n[`Sequential` model](https://www.tensorflow.org/guide/keras/sequential_model),\nwhich is a linear stack of layers. For more complex architectures, you can\neither use the\n[Keras functional API](https://www.tensorflow.org/guide/keras/functional_api),\nwhich lets you build arbitrary graphs of layers, or\n[use subclassing to write models from scratch](https://www.tensorflow.org/guide/keras/making_new_layers_and_models_via_subclassing).\n\nThe `tf.keras.Model` class features built-in training and evaluation methods:\n\n- `tf.keras.Model.fit`: Trains the model for a fixed number of epochs.\n- `tf.keras.Model.predict`: Generates output predictions for the input samples.\n- `tf.keras.Model.evaluate`: Returns the loss and metrics values for the model; configured via the `tf.keras.Model.compile` method.\n\nThese methods give you access to the following built-in training features:\n\n- [Callbacks](https://www.tensorflow.org/api_docs/python/tf/keras/callbacks). You can leverage built-in callbacks for early stopping, model checkpointing, and [TensorBoard](https://www.tensorflow.org/tensorboard) monitoring. You can also [implement custom callbacks](https://www.tensorflow.org/guide/keras/writing_your_own_callbacks).\n- [Distributed training](https://www.tensorflow.org/guide/keras/distributed_training). You can easily scale up your training to multiple GPUs, TPUs, or devices.\n- Step fusing. With the `steps_per_execution` argument in `tf.keras.Model.compile`, you can process multiple batches in a single `tf.function` call, which greatly improves device utilization on TPUs.\n\nFor a detailed overview of how to use `fit`, see the\n[training and evaluation guide](https://www.tensorflow.org/guide/keras/training_with_built_in_methods).\nTo learn how to customize the built-in training and evaluation loops, see\n[Customizing what happens in `fit()`](https://www.tensorflow.org/guide/keras/customizing_what_happens_in_fit).\n\n### Other APIs and tools\n\nKeras provides many other APIs and tools for deep learning, including:\n\n- [Optimizers](https://keras.io/api/optimizers/)\n- [Metrics](https://keras.io/api/metrics/)\n- [Losses](https://keras.io/api/losses/)\n- [Data loading utilities](https://keras.io/api/data_loading/)\n\nFor a full list of available APIs, see the\n[Keras API reference](https://keras.io/api/). To learn more about other Keras\nprojects and initiatives, see\n[The Keras ecosystem](https://keras.io/getting_started/ecosystem/).\n\nNext steps\n----------\n\nTo get started using Keras with TensorFlow, check out the following topics:\n\n- [The Sequential model](https://www.tensorflow.org/guide/keras/sequential_model)\n- [The Functional API](https://www.tensorflow.org/guide/keras/functional)\n- [Training \\& evaluation with the built-in methods](https://www.tensorflow.org/guide/keras/training_with_built_in_methods)\n- [Making new layers and models via subclassing](https://www.tensorflow.org/guide/keras/custom_layers_and_models)\n- [Serialization and saving](https://www.tensorflow.org/guide/keras/save_and_serialize)\n- [Working with preprocessing layers](https://www.tensorflow.org/guide/keras/preprocessing_layers)\n- [Customizing what happens in fit()](https://www.tensorflow.org/guide/keras/customizing_what_happens_in_fit)\n- [Writing a training loop from scratch](https://www.tensorflow.org/guide/keras/writing_a_training_loop_from_scratch)\n- [Working with RNNs](https://www.tensorflow.org/guide/keras/rnn)\n- [Understanding masking \\& padding](https://www.tensorflow.org/guide/keras/masking_and_padding)\n- [Writing your own callbacks](https://www.tensorflow.org/guide/keras/custom_callback)\n- [Transfer learning \\& fine-tuning](https://www.tensorflow.org/guide/keras/transfer_learning)\n- [Multi-GPU and distributed training](https://www.tensorflow.org/guide/keras/distributed_training)\n\nTo learn more about Keras, see the following topics at\n[keras.io](http://keras.io):\n\n- [About Keras](https://keras.io/about/)\n- [Introduction to Keras for Engineers](https://keras.io/getting_started/intro_to_keras_for_engineers/)\n- [Introduction to Keras for Researchers](https://keras.io/getting_started/intro_to_keras_for_researchers/)\n- [Keras API reference](https://keras.io/api/)\n- [The Keras ecosystem](https://keras.io/getting_started/ecosystem/)"]]
