3

I am developing a multi-class perceptron algorithm and was wondering if there are any datasets that could be used to test a multi-class perceptron? - A dataset where the classes are linearly separable and have at least 100 or more instances for training?

Henrik
  • 13,314
  • 9
  • 63
  • 123
KungPaoChicken
  • 321
  • 1
  • 6

1 Answers1

2

Maybe the good old iris? It suits your needs and is good for start.

  • 1
    i think only one of the classes in the Iris data set is linearly separable. (From the OP's Question, i think he's after data sets w/ *only* linearly separable classes). – doug Jul 27 '10 at 09:52
  • @doug, good point, still I don't think that one can get any non-synthetic fully linearly separable problem. Nevertheless I'll wait for the OP's reaction. –  Jul 27 '10 at 10:13
  • yes - only one of the classes in the iris data set is linearly separable - so, I've now started experimenting with algorithmically generated datasets. – KungPaoChicken Jul 27 '10 at 23:22