7

I was reading the paper on Relational Fisher Kernel which involves Bayesian Logic Programs to calculate the Fisher score and then uses SVM to obtain the class labels for each data item.

I don't have strong background from Machine learning. Can someone please let me know about how to go about implementing an end-to-end Relational Fisher Kernel and what sort of input would it expect? I could not find any easy step-by-step flow showing this implementation. I am ok with using libraries for SVM etc. (e.g. libsvm), but I would like to know the end-to-end flow (in as easy language as possible). Any help will be highly appreciated.

Salil
  • 1,739
  • 2
  • 15
  • 25
  • +1 for pointing out an interesting and well written paper – fairidox Nov 23 '11 at 08:35
  • So far, I have deduced that the input will be in the clausal form as in case of Logic Programming. One can use Bayesian Logic Programs (e.g. tool like Balios) to induce a Bayesian N/w. The parameters of the induced n/w are in the form of Conditional Probability Distributions for the clauses (derived using EM in case of Balios). After this, using the formula mentioned in the paper mentioned in the question, one can compute the gradient, and subsequently the kernel. The kernel can be fit into the SVM (e.g. libsvm) and you are done. Can someone please confirm if I am on right track? – Salil Nov 24 '11 at 08:50
  • 1
    You may want to check out this SO clone for machine learning: http://metaoptimize.com/qa – fairidox Nov 24 '11 at 16:11
  • Thanks... perhaps I will try my luck there :-) – Salil Nov 24 '11 at 16:44
  • 7
    @justaname we have a machine learning site on the stackexchange-network: stats.stackexchange.com :(. – steffen Dec 20 '11 at 15:19

1 Answers1

1

libsvm does not implement the Relation Fisher Kernel, however, you can calculate the Fisher information matrix as described in the paper, and the use it as the precomputed kernel input to libsvm. See: using precomputed kernels with libsvm

Community
  • 1
  • 1
Diego
  • 18,035
  • 5
  • 62
  • 66
  • Right, that's what I figured out earlier (in the comments): "using the formula mentioned in the paper, one can compute the gradient, and subsequently the kernel. The kernel can be fit into the SVM (e.g. libsvm)". Thanks for confirming that. – Salil Jun 06 '15 at 03:35