#coding=utf-8
import numpy as np
from random import shuffle
import sklearn
from sklearn import svm
from sklearn import tree
from sklearn.linear_model import LogisticRegression
from sklearn import cross_validation
from sklearn.cross_validation import cross_val_score
from sklearn.cross_validation import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import AdaBoostClassifier
import nltk
from sklearn.metrics import precision_score,recall_score,f1_score,cohen_kappa_score
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.cross_validation import train_test_split

# f=open('d:/pymeng/helpful_paper/final_code/kappa_test.txt')
# file=f.readlines()
# feature_all=np.loadtxt(file)
# f.close()
# tag1=feature_all[:,0]
# tag2=feature_all[:,1]
# print 'kappa value is  %f' % cohen_kappa_score(tag1,tag2)

f=open('d:/pymeng/helpful_paper/final_code/all_feature_value.txt')
file=f.readlines()
feature_all=np.loadtxt(file)
f.close()

f_selected_X=feature_all[:,8:-1] #初始训练集，第8到最后列
f_selected_y=feature_all[:,-1] #标签，不变
f_selected_initial_train,f_selected_initial_test,train_tag,testtag=train_test_split(f_selected_X,f_selected_y,test_size=0.20,random_state=531)
gbdt=GradientBoostingClassifier(n_estimators=23,max_features=None,min_samples_split=1,random_state=0)  
gbdt.fit(f_selected_initial_train,train_tag)
accuracy=precision_score(testtag,gbdt.predict(f_selected_initial_test))
print "base accuracy is %f" %accuracy
print  "base recall rate is %f" %recall_score(testtag,gbdt.predict(f_selected_initial_test))
print "base F-measure is %f"  %f1_score(testtag,gbdt.predict(f_selected_initial_test))


f_selected_base=f_selected_X
f_candidate_all=feature_all[:,:8]
for i in range(8):
	f_selected_i=np.column_stack([f_selected_base,f_candidate_all[:,i]])
	f_selected_initial_train,f_selected_initial_test,train_tag,testtag=train_test_split(f_selected_i,f_selected_y,test_size=0.20,random_state=531)
	gbdt=GradientBoostingClassifier(n_estimators=23,max_features=None,min_samples_split=1,random_state=0)  
	gbdt.fit(f_selected_initial_train,train_tag)
	accuracy=precision_score(testtag,gbdt.predict(f_selected_initial_test))
	print "%d's accuracy is %f" %(i,accuracy)
	print  " recall rate is %f" %recall_score(testtag,gbdt.predict(f_selected_initial_test))
	print "F-measure is %f"  %f1_score(testtag,gbdt.predict(f_selected_initial_test))
	f_selected_base=f_selected_i
	if i==7:break


f_candidate_all=feature_all[:,:8]
a,b=f_candidate_all.shape
f_selected_X=feature_all[:,8:-1] #初始训练集，第8到最后列
f_selected_y=feature_all[:,-1] #标签，不变
f_selected_initial_train,f_selected_initial_test,train_tag,testtag=train_test_split(f_selected_X,f_selected_y,test_size=0.20,random_state=531)
gbdt=GradientBoostingClassifier(n_estimators=23,max_features=None,min_samples_split=1,random_state=0)  
gbdt.fit(f_selected_initial_train,train_tag)
E_selected=f1_score(testtag,gbdt.predict(f_selected_initial_test))

print a,b
while (1):
	E_can_all={}
	for i in range(b):
		fi=f_candidate_all[:,i]
		f_selected_X_fi=np.column_stack([f_selected_X,fi])
		f_selected_train,f_selected_test,train_tag,testtag=train_test_split(f_selected_X_fi,f_selected_y,test_size=0.20,random_state=531)
		gbdt=GradientBoostingClassifier(n_estimators=23,max_features=None,min_samples_split=1,random_state=0)  
		gbdt.fit(f_selected_train,train_tag)
		E_selected_i=f1_score(testtag,gbdt.predict(f_selected_test))
		print "E_selected_%d is %f" %(i,E_selected_i)
		E_can_all[i]=E_selected_i
	E_all_max=max(E_can_all.values())
	E_all_max_index=E_can_all.values().index(E_all_max)
	print "E_all_max_index %d " %E_all_max_index
	if E_all_max>=E_selected:
		fmax=f_candidate_all[:,E_all_max_index]
		f_selected_X=np.column_stack([f_selected_X,fmax])
		E_selected=E_all_max
		f_candidate_all=np.delete(f_candidate_all,E_all_max_index,1)
	a,b=f_candidate_all.shape
	print "numpy column is %d" %b
	print "repeat once again"
	if b==0 or E_all_max<E_selected:break

f_candidate_all=feature_all[:,:8]
f1=f_candidate_all[:,0]
f2=f_candidate_all[:,1]
f3=f_candidate_all[:,4]
f4=f_candidate_all[:,2]
f5=f_candidate_all[:,3]

f1_f2=np.column_stack([f1,f2])
f1_f2_f3=np.column_stack([f1_f2,f3])
f1_f2_f3_f4=np.column_stack([f1_f2_f3,f4])
f1_f2_f3_f4_f5=np.column_stack([f1_f2_f3_f4,f5])


f_selected_base=feature_all[:,8:-1]
for i in range(5):
	f_selected_i=np.column_stack([f_selected_base,f1_f2_f3_f4_f5[:,i]])
	f_selected_initial_train,f_selected_initial_test,train_tag,testtag=train_test_split(f_selected_i,f_selected_y,test_size=0.20,random_state=531)
	gbdt=GradientBoostingClassifier(n_estimators=23,max_features=None,min_samples_split=1,random_state=0)  
	gbdt.fit(f_selected_initial_train,train_tag)
	accuracy=precision_score(testtag,gbdt.predict(f_selected_initial_test))
	print " accuracy is %f" %accuracy
	print  " recall rate is %f" %recall_score(testtag,gbdt.predict(f_selected_initial_test))
	print " F-measure is %f"  %f1_score(testtag,gbdt.predict(f_selected_initial_test))
	f_selected_base=f_selected_i

f_selected=np.column_stack([feature_all[:,8:-1],f1_f2_f3_f4_f5[:,:3]])

f_selected_initial_train,f_selected_initial_test,train_tag,testtag=train_test_split(f_selected,f_selected_y,test_size=0.20,random_state=531)
print "hello"
# 构造分类器
dt=tree.DecisionTreeClassifier()  #决策树
rf=RandomForestClassifier(n_estimators=23)
est=gbdt=GradientBoostingClassifier(n_estimators=23,max_features=None,min_samples_split=1,random_state=0)  



# 训练模型
dt.fit(f_selected_initial_train,train_tag)
rf.fit(f_selected_initial_train,train_tag)
est.fit(f_selected_initial_train,train_tag)


# 预测
dt_predict=dt.predict(f_selected_initial_test)
rf_predict=rf.predict(f_selected_initial_test)
est_predict=est.predict(f_selected_initial_test)


# 分类准确率
print  'dt average accuracy rate is %f' %precision_score(testtag,dt_predict)
print  'rf average accuracy rate is %f' %precision_score(testtag,rf_predict)
print  'est average accuracy rate is %f' %precision_score(testtag,est_predict)


# 分类召回率
print  'dt recall rate is %f' %recall_score(testtag,dt_predict)
print  'rf recall rate is %f' %recall_score(testtag,rf_predict)
print  'est recall rate is %f' %recall_score(testtag,est_predict)

# 分类调和值
print  'dt f-measure is %f' %f1_score(testtag,dt_predict)
print  'rf f-measure is %f' %f1_score(testtag,rf_predict)
print  'est f-measure is %f' %f1_score(testtag,est_predict)