See #13541
Seems that we've decided that all parameters should have a default. (Can someone tell me why? I don't like it.) For some parameters, the default value makes little sense (e.g., n_clusters in KMeans, eps in DBSCAN), I think we should have a way to recommend users to set these parameters manually.
My proposal is to define a DefaultMeaninglessWarning and raise it when these parameters are not set manually.