0
$\begingroup$

Dataset for User-Page view, which represent in matrix, rows are Users (2000+ or more) and columns ( Pages may be 100+),entries represent frequency(number of time user visit that page), I would like to know what type of MDS algorithm should i used ( classical, metric, non-metric etc..)

$\endgroup$

1 Answer 1

0
$\begingroup$

It depends on the dataset and application, do you have any more information on them? (Sorry I would have commented but I can't)

I had a similar application for a TF-IDF matrix, with ~10,000 rows and 76 columns. I had to get a relevance score. For me, metric MDS gave the best results. I used Python with this function.

Since your dataset is not that large, you could try all methods. If that takes too much time then use a subset of the data. Using a subset of data you could also manually check the MDS result to see if it's giving results that are reasonable. Using PCA to verify on the whole dataset is also an option though it's not that reliable.

$\endgroup$
3
  • $\begingroup$ thank you for your note, so do you think t-SNE tool can be good or i have to test MDS, there ar many version in MDS such metric, classical, non-metric, Sammon etc. $\endgroup$
    – Ray ben
    Aug 21, 2016 at 15:41
  • $\begingroup$ I have never used t-SNE but from the paper it looks like a good starting point. I think there are instances where MDS might perform better. You can check this to see what suits your data. If they run fast you could take a vote between them. scikit-learn.org/stable/auto_examples/manifold/… $\endgroup$ Aug 22, 2016 at 12:38
  • $\begingroup$ thanks, can you please visit following link with kind regards, stats.stackexchange.com/questions/231082/… $\endgroup$
    – Ray ben
    Aug 22, 2016 at 12:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.