Vertex AI Vizier は、複雑な機械学習(ML)モデルでのハイパーパラメータのチューニングを支援するブラックボックス最適化サービスです。ML モデルに異なるハイパーパラメータが多数ある場合、手動でのチューニングは難しく時間がかかります。Vertex AI Vizier を使用すると、ハイパーパラメータが調整されてモデルの出力が最適化されます。
Vertex AI Vizier は、ML モデルのハイパーパラメータを最適化しますが、他の最適化タスクを実行することもできます。
パラメータを調整する
Vertex AI Vizier を使用すると、関数内のパラメータを効果的にチューニングできます。たとえば、Vertex AI Vizier を使用して、ニュースサイトの定期購入ボタンの背景色、フォントサイズ、リンクの色の最も効果的な組み合わせを決定します。その他の例については、ユースケースをご覧ください。
Vertex AI Vizier は、多数のパラメータを含む複雑なモデルを最適化するための独立したサービスです。これは ML と ML 以外の両方のユースケースに使用できます。トレーニング ジョブや他のシステム(マルチクラウドも含む)で使用できます。カスタム トレーニングのハイパーパラメータ チューニングは、トレーニング ジョブに Vertex AI Vizier を使用する組み込み機能です。ML モデルに最適なハイパーパラメータ設定を決定します。
ユースケース
次のシナリオで、Vertex AI Vizier はハイパーパラメータを調整してモデルを最適化するか、パラメータを調整して結果を最適化します。
[[["わかりやすい","easyToUnderstand","thumb-up"],["問題の解決に役立った","solvedMyProblem","thumb-up"],["その他","otherUp","thumb-up"]],[["わかりにくい","hardToUnderstand","thumb-down"],["情報またはサンプルコードが不正確","incorrectInformationOrSampleCode","thumb-down"],["必要な情報 / サンプルがない","missingTheInformationSamplesINeed","thumb-down"],["翻訳に関する問題","translationIssue","thumb-down"],["その他","otherDown","thumb-down"]],["最終更新日 2024-12-23 UTC。"],[],[],null,["# Vertex AI Vizier overview\n\nVertex AI Vizier is a tool for optimizing any system with configurable\nparameters where evaluating any given parameter settings is an expensive task.\nWhen ML models have many different hyperparameters, it can\nbe difficult and time consuming to tune them manually. Vertex AI Vizier\noptimizes your model's output by tuning the hyperparameters for\nyou.\n\n*Black-box optimization* is the optimization of a system that meets\neither of the following criteria:\n\n- Doesn't have a known [objective\n function](https://developers.google.com/machine-learning/glossary#objective-function)\n to evaluate.\n\n- Is too costly to evaluate by using the objective function, usually due to\n the complexity of the system.\n\nAdditional Vertex AI Vizier functionality\n-----------------------------------------\n\nVertex AI Vizier optimizes hyperparameters of ML models, but it can also\nperform other optimization tasks.\n\n### Tune parameters\n\nYou can use Vertex AI Vizier to effectively tune parameters\nin a function. For example, use Vertex AI Vizier to determine the\nmost effective combination of background color, font size, and\nlink color on a news website's Subscription button. For more examples,\nsee the [use cases](#use-cases).\n\n[Read about the difference between hyperparameters and\nparameters](/vertex-ai/docs/training/hyperparameter-tuning-overview).\n\n### Optimize any evaluable system\n\nVertex AI Vizier works with any\nsystem that you can evaluate, including systems\nthat can't be expressed as a closed-form analytical function.\nFor example, use\nVertex AI Vizier to find the best neural network\ndepth, width, and learning rate for a TensorFlow\nmodel.\n\nHow Vertex AI Vizier works\n--------------------------\n\nThe following sections define terms, behavior, and available values that you\ncan use with Vertex AI Vizier to optimize your ML model or function.\nStart by determining a [study\nconfiguration](/vertex-ai/docs/reference/rest/v1/projects.locations.studies).\n\n### Study configurations\n\nA *study configuration* is the definition of the optimization problem\nthat you are trying to solve. It includes the\nresult you would like to optimize and the\nhyperparameters or parameters that affect that result.\n\n### Studies and trials\n\nA *study* is the implementation of a study configuration. A\nstudy uses the study configuration's goals (metrics) and input values\n(hyperparameters or parameters) to conduct experiments,\ncalled trials. A *trial* is a specific set of input values that produce a\nmeasured outcome relative to your goals.\n\nVertex AI Vizier suggests input values to use for each trial\nbut it does not run trials for you.\n\nA study continues\nuntil it reaches a set limit of trials, or you interrupt the study. A trial\ncontinues\nuntil you indicate that it is either finished or infeasible.\n\n### Measurements\n\nA *measurement* is the measured outcome of your trial. Each measurement can\ncontain one or more metrics, and each trial can contain one or more\nmeasurements taken over a period of time. You can add a new measurement to\nthe trial at any point before the trial is completed.\n\n### Search algorithms\n\nIf you don't specify an algorithm, Vertex AI Vizier uses the default\nalgorithm. The default algorithm applies Bayesian optimization to arrive at\nthe optimal solution with a more effective search over the parameter space.\n\nThe following values are available:\n\n- `ALGORITHM_UNSPECIFIED`: Same as not specifying an algorithm.\n Vertex AI chooses the best search algorithm between Gaussian\n process bandits, linear combination search, or their variants.\n\n- `GRID_SEARCH`: A simple grid search within the feasible space. This option is\n useful if you want to specify a quantity of trials that is\n greater than the number of points in the feasible space. In such cases, if\n you do not specify a grid search, the default\n algorithm can generate duplicate suggestions. To use grid search, all\n parameters must be of type `INTEGER`, `CATEGORICAL`, or `DISCRETE`.\n\n- `RANDOM_SEARCH`: A simple random search within the feasible space.\n\nHow Vertex AI Vizier differs from custom training\n-------------------------------------------------\n\nVertex AI Vizier is an independent service for optimizing complex models\nwith many parameters. It can be used for both ML and non-ML use cases. It can\nbe used with Training jobs or with other systems (even multicloud).\n[Hyperparameter tuning for custom\ntraining](/vertex-ai/docs/training/hyperparameter-tuning-overview) is a built-in feature\nthat uses Vertex AI Vizier for training jobs. It helps determine the best hyperparameter\nsettings for an ML model.\n\nUse cases\n---------\n\nIn the following scenarios, Vertex AI Vizier helps tune hyperparameters\nto optimize a model or tune parameters to optimize an outcome:\n\n- Optimize the learning rate, batch size, and other hyperparameters\n of a neural network recommendation engine.\n\n- Optimize usability of an application by testing different arrangements of\n user interface elements.\n\n- Minimize computing resources for a job by identifying an ideal buffer size\n and thread count.\n\n- Optimize the amounts of ingredients in a recipe to produce the\n most delicious version.\n\nWhat's next\n-----------\n\n- To learn more about how Vertex AI Vizier tunes multi-objective functions, see [Random Hypervolume Scalarizations for Provable Multi-Objective Black Box Optimization](https://arxiv.org/abs/2006.04655)."]]