Int Prediction

JorgeJorge Member Posts: 19 Maven
edited November 2018 in Help
Hi!

In my new project, I have 5 or 6 nominal attributes and I want to learn with the training example some % valorations (the training example only have ~50-60 of 100 possible % results).

I'm searching a learner to predict int (%) values. Are there any operator with that feature?

Thanks a lot.

Jorge

Answers

  • landland RapidMiner Certified Analyst, RapidMiner Certified Expert, Member Posts: 2,531 Unicorn
    Hi Jorge,
    you could transform this problem into a regression problem by changing the label into a numerical attribute. The regression learners predict a number and not a discret number of classes, so even if a value is not included, it can be returned.
    Most regression learners do not cope with nominal values, so you might have to binomalise them.

    Greetings,
      Sebastian
  • JorgeJorge Member Posts: 19 Maven
    Thanks Sebastian,

    But the regression learners (as far I know) make operations with the values of the atributtes, and they give different results if the transformation to binomial values is different.

    An example:

    attr1: a --> 0
    attr1: b --> 1
    attr1: c --> 2

    gives result 1

    and:

    attr1: a --> 2
    attr1: b --> 1
    attr1: c --> 0

    gives result 2

    And result 1, result 2 are differents

    Am I wrong?

    Thanks another time,
    Jorge
  • keithkeith Member Posts: 157 Maven
    I think what Sebastian meant is to create binomial variables for each possible value of an attribute that is 1 (TRUE) if the attribute is that value, and 0 (FALSE) otherwise.

    e.g.

    if attr1 can have values a, b, or c, then you create three variables:

    attr1_is_a = 0/1
    attr1_is_b = 0/1
    attr1_is_c = 0/1

    So if you have three rows, each with different values for attr1, the three new variables would take on values of:

    attr1 : { attr1_is_a, attr1_is_b, attr1_is_c }
    a : { 1, 0, 0 }
    b : { 0, 1, 0 }
    c : { 0, 0, 1 }

    Hope this helps.

    Keith
Sign In or Register to comment.