Scale invariance is well motivated in the sense that perhaps you already normalized your data (usually to fit [-1, 1] or something) and you would be bummed to d

Scale invariance is well motivated in the sense that perhaps you already normalized your data (usually to fit [-1, 1] or something) and you would be bummed to discover it fucked up your clusters Or likewise: if you have physical data recorded in some units (say, meters), it would suck if the clusters changed if you had measured stuff in another unit instead