2 ms·
Stop using K-means. HDBScan in its GPU implementation in CuML is probably superior, yes even on your "very large" datasets (A100s can be rented for less than 1$
by Der_Einzige 2y ago
Stop using K-means. HDBScan in its GPU implementation in CuML is probably superior, yes even on your "very large" datasets (A100s can be rented for less than 1$ an hour).
- hansvm 2y agoK-means is optimal for some problems, like in bolt quantization (and hierarchical initializations tend to converge quickly). It's not really applicable to TFA, since you'd choose the cluster count via metrics which have nothing to do with elbow criteria or whatever, but the lesson isn't to not use k-means or to use some other clustering algorithm; it's to know your goals and which model best achieves those.
- antman 2y agoTwo different beasts, both can give good results depending on the dataset