AI software is still software. Software engineers need better tools to make better use of AI software. For example, for software defect prediction and software text mining, the default tunings for software analytics tools can be improved with "hyperparameter optimization" tools that decide (e.g.,) how many trees are needed in a random forest. Hyperparameter optimization is unnecessarily slow when optimizers waste time exploring redundant options (i.e., pairs of tunings with indistinguishably different results). By ignoring redundant tunings, the Dodge(E) hyperparameter optimization tool can run orders of magnitude faster, yet still find better tunings than prior state-of-the-art algorithms (for software defect prediction and software text mining).