Feature Engineering for Predictive Modeling Using Reinforcement Learning

Abstract

Feature engineering is a crucial step in the process of predictive modeling. It involves the transformation of given feature space, typically using mathematical functions, with the objective of reducing the modeling error for a given target. However, there is no well-defined basis for performing effective feature engineering. It involves domain knowledge, intuition, and most of all, a lengthy process of trial and error. The human attention involved in overseeing this process significantly influences the cost of model generation. We present a new framework to automate feature engineering. It is based on performance driven exploration of a transformation graph, which systematically and compactly captures the space of given options. A highly efficient exploration strategy is derived through reinforcement learning on past examples.

Physics-constrained Automatic Feature Engineering for Predictive Modeling in Materials Science

“…While it avoids overfitting, to which deep Learning based FE methods are amenable, and improves the efficiency by selecting subsets of engineered features according to stability and information gain, it does not directly produce intepretable features. Khurana et al (Khurana et al 2016) introduced Cognito, which formulates the feature engineering problem as a search on the transformation tree with an incremental search strategy to explore the prominent features and later extended the framework by combining RL with a linear functional approximation (Khurana, Samulowitz, and Turaga 2018) to improve the efficiency. A similar framework has recently been developed by Zhang et al (Zhang et al 2019), who also used a tree-like transformation graph with the DRL policy.…”

Efficient End-to-End AutoML via Scalable Search Space Decomposition

“…Automating Individual Components. Apart from end-to-end AutoML, many efforts have been devoted to studying sub-problems in AutoML: (1) feature engineering [44,42,41,68,43], (2) algorithm selection [82,46,22,19,63,53], and (3) hyper-parameter tuning [32,79,7,51,36,21,57,80,45,39,70,30,76,90,37]. Meta-learning methods [89,26,23] for hyper-parameter tuning can leverage auxiliary knowledge acquired from previous tasks to achieve faster optimization.…”

Dataset Evolver: An Interactive Feature Engineering Notebook

“…Moreover, the implementation of LFE integrated with DATASET EVOLVER has the average response time of one millisecond for recommending transformations per feature. Cognito provides a search-based automation to feature engineering (Khurana et al 2016;Khurana, Samulowitz, and Turaga 2018). It is based on the hierarchical exploration of a transformation tree which is a directed acyclic graph, where nodes are versions of a given dataset and edges are transformations.…”