In the Apache spark API, they have something like below -
// Evaluate clustering by computing Within Set Sum of Squared Errors
val WSSSE = clusters.computeCost(parsedData)
println(s"Within Set Sum of Squared Errors = $WSSSE")
// Evaluate clustering by computing Within Set Sum of Squared Errors
val WSSSE = clusters.computeCost(parsedData)
println(s"Within Set Sum of Squared Errors = $WSSSE")
do you have something similar to computeCost above? If not, any pointers as to how we could do this would be really helpful.
Hi @vikasgautam18
We chose to use another stopping criteria which is the distance epsilon under which every prototypes should move less than this threshold in order to consider convergence achieved. Others prefer WSSSE which also require a % threshold. We could imagine to propose user to choose how it decide to stop the algorithm between epsilon or WSSSE, this last is easy to implemant once you have prototypes and assignations clusters, you can implement easily...
Look at the link : https://discuss.analyticsvidhya.com/t/what-is-within-cluster-sum-of-squares-by-cluster-in-k-means/2706/2
We chose to use another stopping criteria which is the distance epsilon under which every prototypes should move less than this threshold in order to consider convergence achieved. Others prefer WSSSE which also require a % threshold. We could imagine to propose user to choose how it decide to stop the algorithm between epsilon or WSSSE, this last is easy to implemant once you have prototypes and assignations clusters, you can implement easily...
Look at the link : https://discuss.analyticsvidhya.com/t/what-is-within-cluster-sum-of-squares-by-cluster-in-k-means/2706/2
whoops sent that a bit early. I saw your answer in
a stackexchange about image segmentation using Mean shift clustering. I'm trying to segment art images (e.g. Oil paintings) and would like to try your approach out. Do you have some pointers with alrogithm of this repository to pick for that task? And do you think this could reasonably work for art as well? Sorry about the three messages, I'm posting this from the gitter website on mobile and don't see how to delete or edit my messages. Best regards, Lukas
Hi @lukasstreit ,
Combine the scalable version (spark ones) of gradient ascent and epsilon-proximity with euclidean distance, prefer LUV space to RGB. I applied this algorithm on image independently, not to compare many pictures between them, then i m curious about its application, let me know if you encounter issues ;)
Combine the scalable version (spark ones) of gradient ascent and epsilon-proximity with euclidean distance, prefer LUV space to RGB. I applied this algorithm on image independently, not to compare many pictures between them, then i m curious about its application, let me know if you encounter issues ;)
I try to follow the examples of the notebook, but I think that notebook was for version 8.4
@Maki_AG_twitter I've checked
http://www.clustering4ever.org/API%20Documentation/0.9.6/#org.clustering4ever.clusterizables.EasyClusterizable$
What is your error ?
EasyClusterizable compagnion object and method apply works in 0.9.6 as previously exposed.http://www.clustering4ever.org/API%20Documentation/0.9.6/#org.clustering4ever.clusterizables.EasyClusterizable$
What is your error ?
Hi, I have been working on an implementation of something called a "Probabilistic Self-Organizing Map" (based on work by Anouar et al. 1997 and Lebbah 2015). I found out about Clustering4Ever while googling for M. Lebbah, the author of the more recent prSOM paper. My implementation is in C++ and Qt, I visualize the training process using OpenGL. My implementation seems to be working, I get similar results to what I have seen in some other work. But when I use a trained prSOM to predict the class of a new observation, I think there may be a bug in my code that is making the prediction incorrect, or possibly overestimating the likelihood in some cases because my prediction doesn't include covariance, it assumes all the neurons are spherical gaussians. Does Clustering4Ever have any interest or plan to release an implementation of the prSOM?
