Options

How to change parameters in LDA?

HeikoeWin786HeikoeWin786 Member Posts: 64 Contributor II
in Help
Hello all,

I am testing the use of LDA in rapidminer. I find the changes for parameters cannot be done anymore. May i know how to enable that?
E.g. I want to change the number of topics or top words per topics.
And, what is the best parameter setting for LDA to play around with hypertuning?

thanks.

Answers

  • Options
    jacobcybulskijacobcybulski Member, University Professor Posts: 391 Unicorn
    You can use Optimize Parameters (Grid), to play with LDA parameters within, then also inside the optimization loop perhaps build some model which relies on topics rather than your normal attributes, measure performance, and log them all to determine what combination of LDA parameters gives the best performance.
  • Options
    HeikoeWin786HeikoeWin786 Member Posts: 64 Contributor II
    @jacobcybulski
    Thanks Jacob, Meaning to say that I need to include Optimize parameters and build a model? Could you kindly please explain a bit further as I am new to rapidminer and not really can visualize. I google it and also not found the solution. But, again, what I am looking for is the parameter setting as attached. Am I using thw wrong operator?

    thanks.
  • Options
    jacobcybulskijacobcybulski Member, University Professor Posts: 391 Unicorn
    OK, let me step back, I focused on your last comment about hyper-parameter tuning, so I assumed you are developing some model, e.g. using the topics as your new attributes to the model, which then can be tuned based on the number and size of topics. However, I can see that you are talking about the LDA hyper-parameters tuning. I am a bit confused about your initial comment which said "I find the changes for parameters cannot be done anymore. May i know how to enable that?", would it be you just need to scroll down your parameters pane? All the necessary parameters, such as number of topics and the number of top words to be used per topics, as well as hyper-parameter optimization are there! If your select that you would like to optimize the LDA just flick the optimization parameter and it will use both alpha and beta heuristics. If you had some independent performance measure of your process goodness, e.g. you were to use the generated topics as predictors of some label (you could try using those generated but these have already been optimized) then you can follow my comment above. Let me know if this answer at least the first questions.
  • Options
    MartinLiebigMartinLiebig Administrator, Moderator, Employee, RapidMiner Certified Analyst, RapidMiner Certified Expert, University Professor Posts: 3,507 RM Data Scientist
    Hi,
    did you show the expert parameters?

    Best,
    Martin
    - Sr. Director Data Solutions, Altair RapidMiner -
    Dortmund, Germany
  • Options
    HeikoeWin786HeikoeWin786 Member Posts: 64 Contributor II
    @jacobcybulski
    Hello Jacob, Yes, this explains me pretty well. thanks a lot.
  • Options
    HeikoeWin786HeikoeWin786 Member Posts: 64 Contributor II
    @mschmitz

    Hello Martin, Yes I found it now the parameter setting. However, when I run it my perplexity score is like 741.237 - Is there something I am doing not right? 

    Thanks.
Sign In or Register to comment.