DataRobotは、単一データセットに対する自動および手動の特徴量の変換、および複数データセットに対する特徴量探索など、複数の特徴量エンジニアリング手法をサポートしています。 See the table below to learn about the feature transformation options in DataRobot.
|Automatic feature transformations||Understand date-type feature transformations generated by DataRobot.||プライマリー||Calculated during EDA1.|
|Interaction-based transformations||Transform features based in interactions within your primary dataset by enabling an advanced option.||プライマリー||Enabled in project and calculated during EDA2.|
|特徴量探索||Perform multi-dataset, interaction-based feature creation.||セカンダリー||Configured in project and calculated during EDA2.|
|Automatic modeling transformations||Understand the automated feature engineering DataRobot performs as part of the modeling process.||すべて||Performed during modeling.|
|Manual feature transformations||Manually transform features in your dataset, including variable type transformations.||プライマリー||Transformed in project.|
|AI Catalog transformations|
|Spark SQLを使用したAIカタログでのデータの準備||AIカタログ内でSpark SQLクエリーデータセットを使用して、データセットを強化、変換、成形、および組み合わせます。|
What is feature engineering?¶
Feature engineering is the process of preparing a dataset for machine learning by changing existing features or deriving new features to improve model performance. DataRobot's Automated Feature Engineering uses AI to accelerate the transformation of data into machine learning assets, allowing you to build better machine learning models in less time.
Feature engineering takes place after data preparation and ingest, and before model building.
During EDA1, DataRobot analyzes and profiles every feature in each dataset—detecting feature types, automatically transforming date-type features, and assessing feature quality.
Before model building, you can take further advantage of DataRobot's Automated Feature Engineering by enabling interaction-based transformations for primary datasets or defining relationships between multiple datasets using Feature Discovery. You can also manually transform features in your dataset, including variable type transformations, with functions.
During EDA2, DataRobot uses these known interactions, or relationships, to discover relevant features for your ML models and automatically transforms them to address the unique requirements of each algorithm in the blueprint library.
After model building, navigate to the Leaderboard and select a model. There are a few places you can view which transformations DataRobot performed for individual models during the modeling process:
|ブループリント||Displays preprocessing, modeling algorithms, and post-processing tasks for the selected model.||Click Describe > Blueprints.|
|データ品質処理レポート||Displays feature and imputation information for supported blueprint tasks.||Click Describe > Data Quality Handling.|
|係数||Allows you to download coefficients and preprocessing information, including feature transformations, for supported model types.||Click Describe > Coefficients and click Export.|