Skip to content

アプリケーション内で をクリックすると、お使いのDataRobotバージョンに関する全プラットフォームドキュメントにアクセスできます。

AIプラットフォームリリース

DataRobotのマネージドAIプラットフォーム向けに毎月発表されているプレビューと一般提供の新機能を記録しています。 サポート終了のお知らせも含まれており、必要に応じて、サポート終了ガイドにリンクしています。

December and January SaaS feature announcements

December 2024 and January 2025

This page provides announcements of newly released features in both December 2024 and January 2025, available in DataRobot's SaaS multi-tenant AI Platform, with links to additional resources. From the release center, you can also access past announcements and Self-Managed AI Platform release notes.

December and January features

次の表は、新機能の一覧です。

目的別にグループ化された機能
名前 NextGen Classic
GenAI
New LLM, Azure OpenAI GPT-4o mini, now available
モデリング
Expanded feature engineering offerings for large datasets
NextGen model Leaderboard reorganization eases insight navigation
Multilabel modeling available in Workbench
RAM limit for scoring increased for custom tasks
Clustering now supported in Classic Composable ML projects
予測とMLOps
Create and monitor geospatial custom metrics
Compliance documentation template support for text generation projects
Notebooks
Debugging now supported in DataRobot Notebooks and codespaces

GenAI

New LLM, Azure OpenAI GPT-4o mini, now available

Azure OpenAI’s GPT-4o mini is now generally available for all subscribed enterprise users and Trial users. GPT-4o miniは、小型モデルカテゴリーの中で最も先進的なモデルであり、低コストかつ低レイテンシーでより幅広いAIアプリケーションを実現します。 GPT-4o mini excels at text and image processing and in the appropriate use cases, should be considered as a replacement for GPT-3.5 Turbo series models. DataRobotで利用可能なLLMの全リストをご覧ください。適切なモデルを選択するための開発者向けドキュメントへのリンクもあります。

モデリング

Expanded feature engineering offerings for large datasets

This deployment brings time-aware predictions with feature transformations to Workbench, allowing you to leverage the benefits of feature engineering with datasets larger than 10GB for time-aware use cases. You can use this methodology in conjunction with time series wrangling and achieve full transparency of the transformation process. Use the modeling parameters to configure how to assign rows and make predictions based on forecast distance. DataRobot then builds separate models for each distance and makes row-by-row predictions.

NextGen model Leaderboard reorganization eases insight navigation

With this deployment, Leaderboard insights for both predictive and time-aware experiments are grouped into tabs, with each tab representing the insight's function. Use search to find specific insights as well as to open multiple insights within a tab at once.

Two new insights have been introduced:

  • Related Assets, which show which assets are linked to the current model.

  • Metric Scores, which provides a single-view listing all partition scores for all metrics.

In addition, four new insights have been ported from DataRobot Classic:

  • ログ
  • モデル情報
  • ダウンロード
  • Eureqa

Multilabel modeling available in Workbench

Predictive modeling now supports multicategorical targets, allowing you to build multilabel modeling experiments. Multilabel modeling, a kind of classification task that allows each row in a dataset to be associated with one, several, or zero labels, provides addition flexibility beyond standard multiclass modeling. When setting up the experiment, you can also configure settings that remove selected labels to reduce model complexity. Once modeling completes, use the Multilabel: Per Label Metrics insight to evaluate models by summarizing per-label metric performance for metrics across different values of the prediction threshold.

RAM limit for scoring increased for custom tasks

Predictions made for models with custom tasks now utilize dynamic memory allocation instead of a fixed memory limit. The maximum limit has been increased from 4GB to 14GB, supporting experimentation with larger models including deep learning-oriented machine learning experimentation. For increased efficiency, memory allocation for a specific custom task will be determined through testing at fit time.

Clustering now supported in Composable ML projects

Clustering, an application of unsupervised learning that lets you explore your data by grouping and identifying natural segments, is now a supported project type for applying Composable ML to customize blueprints.

予測とMLOps

Create and monitor geospatial custom metrics

When you create custom metrics and hosted custom metrics jobs, you can specify that a metric is geospatial and select a geospatial segment attribute. After you add a geospatial custom metric to a deployment, you can review the metric data on the Custom metrics tab, using the new geospatial metric chart view:

Compliance documentation template support for text generation projects

With this release, users with template administrator permissions can build compliance documentation templates for text generation projects:

For more information, see Generate compliance documentation and Template Builder for compliance reports.

Notebooks

Debugging now supported in DataRobot Notebooks and codespaces

DataRobot Notebooks now offer built-in support for Python debugger (pdb) and IPython debugger (ipdb) to interactively debug your Python code. Choose to activate the debugger before executing code using ipdb.set_trace(), or retroactively debug after an exception occurs in the executed code with %debug magic. You can also debug Python scripts in a codespace from the integrated terminal using pdb.


更新しました January 31, 2025