train_data, train_response=prepare_data(data)
test_data,test_response=prepare_data(data1)
#print train_data[0,:],test_data[0,:]
#train_data,test_data= scale_data(train_data.astype(float),test_data.astype(float))
#model=svm.SVC(probability=True)
model=RandomForestClassifier(n_estimators=500, criterion="entropy")
imp=preprocessing.Imputer(missing_values=0,strategy="median",axis=0)
imp.fit(train_data)
train_data=imp.transform(train_data)
test_data=imp.transform(test_data)
#Pipeline([("imputer",preprocessing.Imputer(missing_values=0.0,strategy="median",axis=0)),("forest", RandomForestClassifier(n_estimators=500, criterion="entropy"))])
#model=ExtraTreesClassifier(n_estimators=500,criterion="entropy")
print '-------find optimze hyperparameter--------'
model.C=35.33 #hyperparam(train_data,train_response,model)
print '---- doing cv now ------------'
cv_score= cv_loop(train_data, train_response, model)
print 'cv mean AUC score= ',cv_score
print '------ fitting full model -------------'
model.fit(train_data, train_response)
preds=model.predict_proba(test_data)[:,1]
#print model.accuracy_score(test_scaled, test_response)
print 'test data AUC score= ', metrics.roc_auc_score(test_response, preds)
print '------------------------------'
print 'entries in test data= ',len(preds)
ct=0
for pred in preds:
#print pred
if pred>=0.5:
ct+=1
importantidx = [t[1]+1 for t in mylist[:impcount]]
Xtr = data[:tr,importantidx]
Ytr = data[:tr,0]
Xtes = data[tr:,importantidx]
Ytes = data[tr:,0]
#### New SVM with selected Features from
for k in K:
kf = KFold(len(Xtr),n_folds=5, shuffle=True)
Error =[]
F1score =[]
for train, test in kf:
#clf = svm.SVC(kernel='poly', class_weight={1: 4})
clf = svm.SVC(kernel='poly', degree =1, class_weight={1:4})
clf.C = k
clf.fit(Xtr[train],Ytr[train])
Yhat = clf.predict(Xtr[test])
print sum(abs(Ytr[test]-Yhat))
Error.append(1- abs(Ytr[test]-Yhat).mean())
F1score.append(f1_score(Ytr[test], Yhat, average='weighted'))
scores.append(np.mean(F1score))
scores_std.append(np.std(F1score))
"""Plotting section goes here
I will plot scores +, - one standard deviation"""
plt.figure(1, figsize=(4, 3))
