def train(FIS_name,
data,
target_col,
mf,
Ncentroids,
overlap,
alpha=0.5,
iterations=50,
sa=False,
sa_plot=False):
'''
Trains a FIS, writes all the properties of this FIS to a FIS file
using the write function.
Inputs:
data: nummpy array of size > number of centroids x 2
target_col: integer index of the target column
Ncentroids: either an integer (for each feature te same)
or an array size = number of features
mf: 'triangle', 'trapezoid' or 'Gaussian'
overlap: number between 0 and 1,
when gaussian mf overlap is the variance
when triangle/trapezoid overlap is half of the base
iterations: number of iterations for the simulated annealing
Outputs:
RB: list of lists of integer rules
target_centroids: list with scaled target centroids
feature_centroids: the other feature centroids
'''
# scale the data
data, min_x, max_x = scale(data)
# get centroids
centroids = cluster(data, target_col, Ncentroids, plot=False)
# learn WM rules
RB = WM.learn(data, centroids, overlap, mf, target_col)
# return everything needed for testing
target_centroids = centroids[target_col]
# delete target centroid for testing
feature_centroids = np.delete(centroids, target_col, 0)
# delete target values for testing
targets = data[:, target_col]
data = np.delete(data, target_col, 1)
method = 'WM'
# for simulated annealing, get the new rule base
if sa:
method = 'WM+SA'
RB = SA.search(data,
targets,
RB,
alpha,
feature_centroids,
overlap,
mf,
target_centroids,
min_x[target_col],
max_x[target_col],
plot=sa_plot,
iterations=iterations)
# Write FIS file in the format:
# FIS_name.FIS
with open(FIS_name + '.FIS', "w") as fis_file:
write(fis_file, method, mf, overlap, target_centroids,
feature_centroids, RB)
