class knn_classifier(Classifier):
def __init__(self,ticker,inputSize=5, binary=True, n_neighbors=15, risk=0.5, adaboost=False):
self.type = 'KNN'
self.ticker=ticker
self.days=inputSize
self.inputSize = inputSize
self.binary=binary
self.risk_thresh = risk
self.adaboost = adaboost
if binary:
self.clf = neighbors.KNeighborsClassifier(n_neighbors, weights='distance')
else:
self.clf = neighbors.KNeighborsRegressor(n_neighbors, weights='distance')
if adaboost: self.clf = AdaBoostRegressor(base_estimator=self.clf, n_estimators=100)
def predict(self, inputArray):
inputArray = np.array(inputArray)
inputArray.reshape([1,-1])
if self.binary:
pred = self.clf.predict_proba(inputArray)
pred = (np.array(pred)[:,1] > self.risk_thresh)*1
else:
pred = self.clf.predict(inputArray)
return pred
def fit(self, X, Y):
self.clf.fit(X,Y)
class gaussProcess_classifier(Classifier):
def __init__(self,
ticker,
inputSize=5,
binary=True,
risk=0.5,
numTrainDays=300,
adaboost=False):
self.type = 'Gaussian Process'
self.ticker = ticker
self.days = inputSize
self.inputSize = inputSize
self.binary = binary
self.risk_thresh = risk
self.adaboost = adaboost
self.numTrainDays = numTrainDays
if binary:
self.clf = GaussianProcessClassifier()
else:
self.clf = GaussianProcessRegressor()
if adaboost:
self.clf = AdaBoostRegressor(base_estimator=self.clf,
n_estimators=100)
def trainClf(self, endDay=date.today(), numTrainDays=100):
X, Y = self.processData(endDay, self.numTrainDays)
self.fit(X, Y)
def predict(self, inputArray):
inputArray = np.array(inputArray)
inputArray.reshape([1, -1])
if self.binary:
pred = self.clf.predict_proba(inputArray)
pred = (np.array(pred)[:, 1] > self.risk_thresh) * 1
else:
pred = self.clf.predict(inputArray)
return pred
def fit(self, X, Y):
self.clf.fit(X, Y)
class dt_class(Classifier):
def __init__(self,
ticker,
inputSize=5,
binary=True,
risk=0.5,
adaboost=False):
self.type = 'Decision Tree'
self.ticker = ticker
self.days = inputSize
self.inputSize = inputSize
self.binary = binary
self.adaboost = adaboost
self.risk_thresh = risk
if binary:
self.clf = tree.DecisionTreeClassifier(max_depth=inputSize)
if adaboost:
self.clf = AdaBoostClassifier(base_estimator=self.clf,
n_estimators=100)
else:
self.clf = tree.DecisionTreeRegressor(max_depth=inputSize)
if adaboost:
self.clf = AdaBoostRegressor(base_estimator=self.clf,
n_estimators=100)
def predict(self, inputArray):
inputArray = np.array(inputArray)
inputArray.reshape([1, -1])
if self.binary:
pred = self.clf.predict_proba(inputArray)
pred = (np.array(pred)[:, 1] > self.risk_thresh) * 1
else:
pred = self.clf.predict(inputArray)
return pred
def fit(self, X, Y):
self.clf.fit(X, Y)
class svm_class(Classifier):
def __init__(self,ticker,inputSize=5, binary=True, risk=0.5, adaboost=False):
self.type = 'SVM'
self.ticker=ticker
self.days=inputSize
self.inputSize = inputSize
self.binary=binary
kern = 'sigmoid'
self.risk_thresh = 1 - risk
self.adaboost = adaboost
if binary:
self.clf = svm.SVC(kernel=kern)
self.clf.probability=True
if adaboost: self.clf = AdaBoostClassifier(base_estimator=self.clf, n_estimators=100)
else:
self.clf = svm.SVR(kernel=kern)
if adaboost: self.clf = AdaBoostRegressor(base_estimator=self.clf, n_estimators=100)
def predict(self, inputArray):
inputArray = np.array(inputArray)
inputArray.reshape([1,-1])
if self.binary:
pred = self.clf.predict_proba(inputArray)
pred = (np.array(pred)[:,1] > self.risk_thresh)*1
else:
pred = self.clf.predict(inputArray)
return pred
def fit(self, X, Y):
self.clf.fit(X,Y)
clf.fit(subTrainFeature, subTrainLabel)
predictedTrainProb = clf.predict(trainFeature)
predictedTestProb = clf.predict(testFeature)
for item in predictedTrainProb:
newTrainFeature_temp.append(item)
for item in predictedTestProb:
newTestFeature_temp.append(item)
newTrainFeature.append(newTrainFeature_temp)
newTestFeature.append(newTestFeature_temp)
newTrainFeature = np.array(newTrainFeature).T
newTestFeature = np.array(newTestFeature).T
clf = linear_model.LogisticRegression(penalty='l2', dual=False, class_weight='auto')
clf.fit(newTrainFeature, trainLabel)
predictedLabel = clf.predict_proba(newTestFeature)
return(predictedLabel[:, 0])
if(__name__ == "__main__"):
trainFeature, trainLabel, testFeature, testPlatform = readFeature(5, 0.5, 10, 0.6, 15, 0.6, 5, 0.6, 1)
'''
selectFeature = SelectKBest(chi2, k = 55)
selectFeature.fit(trainFeature, trainLabel)
trainFeature_new = selectFeature.transform(trainFeature)
testFeature_new = selectFeature.transform(testFeature)
'''
trainFeature_new = trainFeature[:, :]
testFeature_new = testFeature[:, :]
'''
trainFeature_new = trainFeature[:, :26]
testFeature_new = testFeature[:, :26]
preds.to_csv('/Users/IkkiTanaka/Documents/KDDCup/pred/xgb/sk_GBM2.csv',
header=None,
index=False)
new_label = a.sort(0).iloc[(a.sort(0)[0] > 0.01).values][1].values
clf = GradientBoostingClassifier(n_estimators=400,
learning_rate=0.05,
subsample=.96,
max_depth=4,
verbose=1,
max_features=.96,
random_state=None)
new_dtrain_sp = dtrain_sp[new_label]
new_dval = dval[new_label]
clf.fit(dtrain_sp, label_dtrain[0].values)
pred = clf.predict_proba(dval)
print("ROC score", metrics.roc_auc_score(label_dval[0].values, pred[:, 1]))
#GaussianNB
from sklearn.naive_bayes import GaussianNB
clf = GaussianNB()
clf.fit(dtrain_sp, label_dtrain[0].values)
pred = clf.predict_proba(dval)
print("ROC score", metrics.roc_auc_score(label_dval[0].values, pred[:, 1]))
scaler = StandardScaler()
dtrain_sp = scaler.fit_transform(dtrain_sp)
dval = scaler.transform(dval)
from sklearn import svm
clf = svm.SVC(C=1,
pred = calibrated_clf.predict_proba(dtest)
sample = pd.read_csv('/Users/IkkiTanaka/Documents/KDDCup/sampleSubmission.csv',header=None)
preds = pd.concat([sample[0],pd.DataFrame(pred[:,1])],axis=1)
preds.to_csv('/Users/IkkiTanaka/Documents/KDDCup/pred/xgb/sk_GBM2.csv' ,header=None,index=False)
new_label = a.sort(0).iloc[(a.sort(0)[0]>0.01).values][1].values
clf = GradientBoostingClassifier(n_estimators=400,learning_rate=0.05,subsample=.96,max_depth=4,verbose=1,max_features=.96, random_state=None)
new_dtrain_sp = dtrain_sp[new_label]
new_dval = dval[new_label]
clf.fit(dtrain_sp, label_dtrain[0].values)
pred = clf.predict_proba(dval)
print("ROC score", metrics.roc_auc_score(label_dval[0].values, pred[:,1]))
#GaussianNB
from sklearn.naive_bayes import GaussianNB
clf = GaussianNB()
clf.fit(dtrain_sp, label_dtrain[0].values)
pred = clf.predict_proba(dval)
print("ROC score", metrics.roc_auc_score(label_dval[0].values, pred[:,1]))
scaler = StandardScaler()
dtrain_sp = scaler.fit_transform(dtrain_sp)
dval = scaler.transform(dval)