Skip to main content

default.greedy

construct_portfolio

def construct_portfolio(regret_matrix, meta_features, regret_bound)

The portfolio construction algorithm.

Reference: Mining Robust Default Configurations for Resource-constrained AutoML.

Arguments:

  • regret_matrix - A dataframe of regret matrix.
  • meta_features - None or a dataframe of metafeatures matrix. When set to None, the algorithm uses greedy strategy. Otherwise, the algorithm uses greedy strategy with feedback from the nearest neighbor predictor.
  • regret_bound - A float of the regret bound.

Returns:

A list of configuration names.