The RandomSubsetElasticNet model will split training data into randomly generated subsets and train an ElasticNet model for each subset. Subsets are determined by splitting on the training data's binary categorical features. These feature indices must be provided to the model at instantiation. Subsets are created with replacement, so one sample can be part of the training data for multiple ElasticNet models. These subsets are created until a percentage of the training data greater than or equal to the coverage_threshold fits at least one generated models to create is provided, it will make boolean subsets until that number is hit. subset. If an explicit number of models to create is provided, it will make boolean subsets until that number is hit (determined by explicit_model_count). It can exit early if no more samples are covered after a certain number of attempts are made (determined by max_boolean_generation_attempts). Each boolean statement must cover between lower_bound and upper_bound percent of the complete training data. There will also be a fallback ElasticNet model which is trained on the entirety of the training data. For each model created, the R^2 score will be recorded fitted against the subset of samples which match the corresponding boolean phrase.
For prediction and scoring, the testing sample will be tested against all models whose corresponding boolean phrase matches. The predictive score is determined by either the best performing matching model, a weighted subset determined by the individual R^2 values, or a mix between the two. This weight is determined by the p value.
RandomSubsetElasticNet uses sklearn.metrics.r2_score and sklearn.linear_model.ElasticNet.