ML Feature Recommender for Automated Model Development
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current feature stores in machine learning require data scientists to have prior knowledge of existing features, making it difficult for them to discover relevant features for their models, especially in real-time production scenarios, where feature engineering accounts for a significant portion of the development time.
Innovation Solution
A machine learning feature recommender system that collects data and recommends features based on user inputs, user behavior, and feature importance, facilitating the discovery of performant features by data scientists and improving model accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If data scientists search for features in a central repository with metadata layer, then feature reuse can be facilitated, but data scientists must understand the features they want which increases the difficulty of feature discovery
Solution Approach 1:
The system automatically generates feature recommendations based on model type, business problem, and historical data without requiring data scientists to manually search or understand existing features. The recommendation engine serves itself by autonomously selecting and suggesting relevant features, eliminating the need for data scientists to have prior knowledge of the feature repository contents.
Solution Approach 2:
An AI-based recommendation engine acts as an intermediary between data scientists and the feature repository. Instead of data scientists directly searching for features requiring their understanding, the recommendation engine translates high-level business problems into specific feature recommendations, bridging the gap between business objectives and technical feature implementation.
2Adaptability or versatility
If feature engineering is performed manually by data scientists, then feature customization can be achieved, but feature engineering accounts for around 80% of development time which reduces overall productivity
Solution Approach 1:
The system performs preliminary feature selection and engineering automatically based on pre-defined models and historical successful implementations. By preparing and pre-selecting features before data scientists need them, the system eliminates the time-consuming manual feature engineering process while maintaining customization through automated parameter adjustment and selection.
3Loss of time
If data scientists rely on their own knowledge to select features, then feature selection can be done quickly, but relevant features may be missed reducing model accuracy
Solution Approach 1:
The recommendation engine incorporates feedback loops that analyze historical model performance, feature usage patterns, and business outcomes to continuously improve feature recommendations. This feedback mechanism ensures that the system learns from past successes and failures, increasingly accurate feature selections over time while maintaining rapid deployment speeds.
Data Source
AI summary
The described technology is generally directed towards a machine learning feature recommender, for use in connection with a feature store. By collecting data and recommending machine learning features to users based on collected data, embodiments can facilitate data scientists' discovery of features that have been used by their colleagues and that are likely to make their machine learning models more performant. The disclosed machine learning feature recommender can reduce the effort involved in developing machine learning models.