HyperBand(configspace=None, eta=3, min_budget=0.01, max_budget=1, **kwargs)¶
Hyperband implements hyperparameter optimization by sampling candidates at random and “trying” them first, running them for a specific budget. The approach is iterative, promising candidates are run for a longer time, increasing the fidelity for their performance. While this is a very efficient racing approach, random sampling makes no use of the knowledge gained about the candidates during optimization.
- configspace (ConfigSpace object) – valid representation of the search space
- eta (float) – In each iteration, a complete run of sequential halving is executed. In it, after evaluating each configuration on the same subset size, only a fraction of 1/eta of them ‘advances’ to the next round. Must be greater or equal to 2.
- min_budget (float) – The smallest budget to consider. Needs to be positive!
- max_budget (float) – the largest budget to consider. Needs to be larger than min_budget! The budgets will be geometrically distributed $sim eta^k$ for $kin [0, 1, … , num_subsets - 1]$.
Hyperband uses SuccessiveHalving for each iteration. See Li et al. (2016) for reference.
Parameters: iteration (int) – the index of the iteration to be instantiated Returns: SuccessiveHalving – corresponding number of configurations Return type: the SuccessiveHalving iteration with the