def UF(X, y):
# 计算feature间的相关性，进行选择，这里做的是forward
k_range = [50 * i for i in range(1, 4)]
for k in k_range:
selector = SelectKBest(chi2, k=k)
X_ = selector.fit_transform(X, y)
X_train, X_test, y_train, y_test = pre_process(X_, y, bReset=True)
SVM_recommend_run(F_UF,
X_train,
X_test,
y_train,
y_test,
paras={'k-best': k})
def SFM(X, y):
# 从模型中选择，根据重要性，类似逐个选择，后向选择，逐渐抛弃不重要的
X_train, X_test, y_train, y_test = pre_process(X, y)
clf = SVM_recommend()
m_range = [2000 - 50 * i for i in range(36, 40)]
for m in m_range:
selector = SelectFromModel(
clf, threshold=-np.inf,
max_features=m) # 只根据max_features确定选择的数量，不设定threshold
X_ = selector.fit_transform(np.asarray(X), np.asarray(y))
X_train, X_test, y_train, y_test = pre_process(X_, y, bReset=True)
clf = SVM_recommend_run(B_SFM,
X_train,
X_test,
y_train,
y_test,
paras={'max-features': m})
def VT(X, y):
# 利用方差进行选拔, 这里是backward
for var in [0.03 * i for i in range(1, 50)]:
selector = VarianceThreshold(threshold=var)
X_ = selector.fit_transform(X)
# X.shape[1] # selected feature amount
X_train, X_test, y_train, y_test = pre_process(X_, y, bReset=True)
# 有问题
SVM_recommend_run(B_VT,
X_train,
X_test,
y_train,
y_test,
paras={
'variance': var,
'feature-num': X_.shape[1]
})
if feature not in good_features:
selected_features = list(good_features) + [feature]
Xts = np.column_stack(X[:, j] for j in selected_features)
score = evaluateScore(Xts, y)
scores.append((score, feature))
print("Current AUC : ", np.mean(score))
good_features.add(sorted(scores)[-1][1])
score_history.append(sorted(scores)[-1])
print("Current Features : ", sorted(list(good_features)))
# Remove last added feature
good_features.remove(score_history[-1][1])
good_features = sorted(list(good_features))
print("Selected Features : ", good_features)
return good_features
def transform(X, y):
good_features = selectionLoop(X, y)
return X[:, good_features]
if __name__ == "__main__":
os.chdir('..')
X, y = load_data_small()
print(X.shape)
X_ = transform(X, y)
X_train, X_test, y_train, y_test = pre_process(X_, y, bReset=True)
SVM_recommend_run(AUC, X_train, X_test, y_train, y_test,
{'feature-num': X_.shape[1]})
import sys
sys.path.append('..')
import os
import numpy as np
from lib.genetic import *
from configs import *
from load_data import load_data_small
from pre_process import pre_process
from baseline import SVM_recommend, SVM_recommend_run
if __name__ == "__main__":
os.chdir('..')
clf = SVM_recommend()
X, y = load_data_small()
for i in range(1,10):
ga_selector = FeatureSelectionGA(clf,X,y,verbose=1)
feature_num = 200*i
pop = ga_selector.generate(feature_num)
X_ = X[:,pop]
X_train, X_test, y_train, y_test = pre_process(X_, y,bReset=True)
SVM_recommend_run(GA, X_train, X_test, y_train, y_test, {'feature_num': feature_num})
print("t-SNE: %.2g sec" % (t1 - t0))
x_min, x_max = X_.min(0), X_.max(0)
X_norm = (X_ - x_min) / (x_max - x_min) # 归一化
plt.figure(figsize=(8, 8))
for i in range(X_norm.shape[0]):
#plt.text(X_norm[i, 0], X_norm[i, 1], str(y[i]), color=plt.cm.Set1(y[i]),
# fontdict={'weight': 'bold', 'size': 9})
plt.scatter(X_norm[i, 0],
X_norm[i, 1],
color=cm(1. * y[i] / NUM_COLORS))
plt.xticks([])
plt.yticks([])
name = str(n_components) + " " + str(perplexity) + " " + str(
random_state) + ".png"
plt.savefig(name)
plt.show()
#X_ = transform(X, y)
X_train, X_test, y_train, y_test = pre_process(X_,
y,
bReset=True)
SVM_recommend_run(tSNE,
X_train,
X_test,
y_train,
y_test,
paras={
'n_cp': n_components,
'ppl': perplexity,
'rd': random_state
})
import sys
sys.path.append('..')
from boruta import Boruta
import os
from configs import *
from baseline import SVM_recommend, SVM_recommend_run
from load_data import load_data
from pre_process import pre_process
if __name__ == "__main__":
os.chdir('..')
clf = SVM_recommend()
feat_selector = BorutaPy(clf, n_estimators='auto')
feat_selector.fit(X, y)
print(feat_selector.support_)
selected = X[:, feat_selector.support_]
print("")
print("Selected Feature Matrix Shape")
print(selected.shape)
X_train, X_test, y_train, y_test = pre_process(X, y)
SVM_recommend_run(BORUTA, X_test, X_test, y_train, y_test,
{'feature_num': selected.shape[1]})