QBSO-FS for regression and selected features?
hanamthang opened this issue · 2 comments
Hello,
Thank you for the sharing of the great work. In the results created by QBSO-FS:
- What is the best solution found in terms of [0,0,1,1]?
- Number of features used : 2: How can I link to what is the selected features?
Must 10% used features : [(2, 100)]
Best solution found : [0, 0, 1, 1]
Accuracy of found sol : 97.33
Number of features used : 2
Size of solutions dict : 16
Average time to evaluate a solution : 2.928 s
Global optimum : 1111, 97.33
Return (Q-value) : 1.1472914218408514
Time elapsed for execution 1 : 47.79 s
And can this one used for a regression problem?
Many thanks,
Thang
Hello,
I am sorry that I did not answer your question earlier, in fact, I just noticed your issue.
Unfortunately, I am not maintaining this code anymore, but to answer your questions:
1- The vector is mapped to each dataset, so [0,0,1,1] ( I assume it is Iris dataset ) means that we keep the last two features, and the 2 is the number of features used, I do not think that it is possible to go back from the number of features only, that is why, if I am not mistaken, we were using a dictionary ( in a Python context, otherwise, you can call it a HashMap in JAVA ), to only keep the indexes of the used features.
2- I do not think that this method could be applied in a straightforward way to regression problems, but you can always try, and I will be curious to see the results.
Amine
Hello hanamthang/amineremache,
In the line: Must 10% used features : [(2, 100)]
What is the meaning of 100?
Why 100 is taken, I don't understand it. Can you please explain?