Table - PMC

Skip to main content

View full-text article in PMC

. 2021 Apr 13;23(4):457. doi: 10.3390/e23040457

Algorithm 1. Relief algorithm

1:
Input: Features: F, time of iteration: m

2:
Initialize weight value of all features to 0, T is the empty set;

3:
Fori = 1 to m, do

4:
Obtain a sample $S_{i}$ randomly from the training data set $S$ ;

5:
Calculate $S_{i}$ ’s nearest neighbor sample point $S_{i N H}$ from homogeneous training sets, and $S_{i}$ ’s nearest neighbor sample point $S_{i N M}$ from $S_{i}$ different kinds of training sets;

6:
Forj = 1 to a, do

W (F_{j}) = W (F_{j}) - d i f f (F_{j}, S_{i}^{F_{j}}, S_{i N H}^{F_{j}}) / m + d i f f (F_{j}, S_{i}^{F_{j}}, S_{i N M}^{F_{j}}) / m

(2)

7:
end for

8:
end for

9:
Forj = 1 to a, do

10:
If $W (F_{j}) \geq δ$

11:
Add feature $F_{j}$ to T

12:
end if

13:
end for

14:
Output: selected feature subset T