3 ms·
>The agglomeration method that we adopt is Ward’s method, which aims to min- imize the total within-cluster variance. At each step in Ward’s method, the algorit
by refactor_master 5y ago
>The agglomeration method that we adopt is Ward’s method, which aims to min- imize the total within-cluster variance. At each step in Ward’s method, the algorithm finds two clusters that will result in minimum increase in total within-cluster variance after fusing. The number of clusters was narrowed by cluster size and similarity to 9 total groups through review of dendrograms and heat maps by non-clinician investigators (A.G. and N.A.).
1. Pick an algorithm.
2. Get a number of clusters. Preferably below 10, because otherwise it gets too messy for publication.
3. Justify algorithm based on its first paragraph on Wikipedia.
- IshKebab 5y ago"study suggests height comes in 9 distinct types!"
- snissn 5y agoi like your spin on it to height. albiet pain likely has a multidimensional graph to cluster onto, height having one dimension makes the whole idea of clustering it silly :)
- specialist 5y agoI'm much more enthusiastic. (see other comment) More over, picking new categories is a great application of "big data". Real example: Imagine catalog for fashion retailer. Time to add color picker & filter. No agreement on color names, how many shades of gray, etc. So one of our big data gurus figures it out. Gather dozens of palettes. Social cognition style guessing game. Slice & dice our product catalog 100s of ways to get best fit. Result was a thing of beauty. Truly novel. The analysis worthy of a PhD.