DISCOVERING GENOMES TO USE IN MACHINE LEARNING TECHNIQUES

摘要

A facility for identifying combinations of feature and machine learningalgorithm parameters, where each combinationcan be combined with one or more machine learning algorithms to train a model,is disclosed. The facility evaluates each genome basedon the ability of a model trained using that genome and a machine learningalgorithm to produce accurate results when applied to avalidation data set by, for example, generating a fitness or validation scorefor the trained model and the corresponding genome used totrain the model. Genomes that produce fitness scores that exceed a fitnessthreshold are selected for mutation, mutated, and the processis repeated. These trained models can then be applied to new data to generatepredictions for the underlying subject matter.