Timeline for Running k-means clustering with k = 2 recursively on clusters greater than a certain size
Current License: CC BY-SA 4.0
5 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jul 16, 2019 at 18:02 | comment | added | Has QUIT--Anony-Mousse | I don't use autoencoders, because they only work on images and I don't use images. Make sure you don't overfit them. | |
| Jul 16, 2019 at 9:48 | comment | added | mkt | @akhetos You should probably post those as separate questions. Comments are not the best place to resolve major new questions. | |
| Jul 16, 2019 at 6:51 | comment | added | akhetos | Even if it's hard to says without looking at the data, do you have any idea of what to do when silouette score isn't a good indicator of clustering quality? Since conv net and autoencodeur are working well, can I says feature extraction is good? | |
| Jul 16, 2019 at 6:45 | comment | added | akhetos | I want to cluster pattern on image. Varibility intra and inter cluster is huge. For my experiment I use a small data set of 3000 images. I do feature extraction with different tools (conv net, autoencodeur, T-SNE...) but it's look like using distance tools is bad for clustering => even the methodology which gives me the best value of silouette score give me poor interpratibility result (I used hierarchical and K mean). My idea of many K mean with k=2 is to extract some informative cluster. I'll probably miss important information, but I expect cluster to give me at least some information | |
| Jul 16, 2019 at 5:54 | history | answered | Has QUIT--Anony-Mousse | CC BY-SA 4.0 |