0

I have a dataset of containing 4000 x 8 double values. I want to apply genetic programming to learn the optimal parameters (a,b,c,d) of the trapezoidal membership function (trapmf) of fuzzy logic toolbox of matlab, corresponding to each data point/row. What I am trying to do here is learn the fuzzy sets corresponding to each row (We assume the min and max of the column as the two boundary points for that input, i.e the constraint would be: min < a < b < c < d < max).

The ultimate objective here is to optimize the 4 parameters of trapmf function. The dataset that I have contains data for 4000 patients.

1 - 1000 - SARS data

1000 - 2000 - Normal Person data

2000 - 3000 - Pneumonia data

3000 - 4000 - BLP data

The columns from 1-4 are an indication of fever related symptoms, 5-6 HBP related, 7 Nausea related and 8 abPain related data. I want to fuzzify the dataset using transmf and get the optimal value of a,b,c and d corresponding to each row of the dataset.

Any suggestions on how to go about the problem?

Rajesh
  • 1
  • 2
  • 1
    Genetic programming? Correct me if I'm wrong, but it seems that you want to evolve 4 real-valued parameters. Where is the *programming* in that? Also please make your task more clear - you want to evolve one fuzzy set that fits all data points? Also, regarding the 4000 x 8 dataset, 4000 of what and 8 of what? Please describe your case in more detail and in a more systematic way. – zegkljan Jan 03 '15 at 16:29
  • @JanŽegklitz I have made some edits to my previous post if it clarifies the situation a bit better. Genetic programming was my first line of thought to learn the parameters of the trapezoidal function. I want to evolve 4000 fuzzy sets corresponding to each row of the matrix (specific to a particular patient) – Rajesh Jan 04 '15 at 08:40
  • @user3068933 I still don't understand. You want one trapezoidal fuzzy set per one datapoint, correct? And what should this set model? You have 8 features. Sorry, I don't know fuzzy things very well so maybe I just miss some basic knowledge, but genetic algorithms/programming is my area of expertise. – zegkljan Jan 04 '15 at 08:49
  • @JanŽegklitz Yes I want one fuzzy set per data point. Please follow this link [trapmf] (http://au.mathworks.com/help/fuzzy/trapmf.html) which shows how trapezoidal curve depends on vector x and four scalar parameters a, b, c, and d. The additional constraint to be satisfied is the the min and max for the columns. So how do I model this using genetic programming ? – Rajesh Jan 04 '15 at 09:03
  • But how does one membership function (i.e. a function that takes one number and returns the membership of this number to the fuzzy set represented by this function) model the 8 features of the data point? – zegkljan Jan 04 '15 at 09:10
  • @JanŽegklitz The objective function should take in an input vector x (containing the 8 features) and return the 4 parameters (a,b,c and d) corresponding to the vector. But the 4 parameters should satisfy the dditional constraint: min < a < b < c < d < max – Rajesh Jan 04 '15 at 09:24
  • Ok, so the membership function should cover the 8 features of the datapoint. How? What is the criterium to optimize? Maybe you could provide a simple example (please edit your original question). – zegkljan Jan 04 '15 at 12:13

0 Answers0