In the next few months we want to start working on an optimization extension to KNIME. This may include genetic algorithms, simulated annealing but also numerical optimization techniques. What kind of nodes could be useful for you? I already have a node that selects subsets of rows from a table based on several criteria on the columns (e.g. maximize sum of values in the subset or maximize distance between all rows in the subset). The wishlist is open :-)
I would welcome a genetic algorithm loop, which is executed for n iterations. for example it can be used to find out good attributes for regressions/class prediction (regression learner, neuronal networks etc.)
perhaps one can implement an interface to Octave / Matlab....
@Jürgen Probably not the right threat for.. its about optimization...
There are some good nodes to call R functions.
Have you tried to call matlab with the external node extension?
Perhaps it's a misunderstanding on my side....
I will check the link to octave/matlab via the external node
Is this still under consideration or development? In our organization the ability to optimize the parameters of a nonlinear equation to fit a set of data would be quite useful (e.g. nonlinear regression learner/predictor nodes).
Perhaps something akin to the 'Solver' functionality in Excel would also be useful (with support for constraints).
P.S. You may be able to leverage the Apache Commons Math optimization package.
I would suggest :
(1) a LP file loader to populate optimizaiton model, with a output to a LP file writer node
(2) a Solver connector to feed solver (since LP file is accepted by most commercial solver), and retrieve result (may be 2-rows-table) from solver.
Currently, I am working on genetic algorithm for feature selection. I tried a couple of methods such as matlab scripts integration, R integration. But I still have not figured it out.
Could you please instruct me how I should do that? Is there any kinme nodes or workflows could accomplish this task?
Thank you very much!
There are currently no nodes for feature selection with a GA. We have a meta node for standard backward feature elimination, though. And the optimization extension contains a GA implementation that can be used in custom nodes.
Thank you very much, Thor! It seems I have to figure it out by myself.