I am trying to predict the solution time for riddles in which matchsticks are combined into digits and operators. An example of a matchstick riddle is 4-2=8. The solution for this riddle would be obtained by moving one matchstick from the ‘8’ to the ‘-’ sign, resulting in the correct equation 4+2=6. The data consists of 100 riddles and the corresponding solution times. The two types of features that are available for each riddle are:

  • a 23 dimensional binary vector that indicates which of the available positions are filled with matches or
  • a 12-dimensional integer vector that counts the appearance of each token (10 digits, 2 operators)

Although today neural nets are very popular I am not sure that a neural net is the best choice for this particular problem. Firstly, because the data set is very small. Secondly because of the binary inputs. What might be a more effective model for this problem ?



You must log in to answer this question.

Browse other questions tagged .