gbr1 = GradientBoostingRegressor(loss='ls', learning_rate=0.1, n_estimators=100, subsample=1.0,
min_samples_split=1, min_samples_leaf=3, max_depth=15, init=None,
random_state=None, max_features=None, alpha=0.9, verbose=2)
# Select Greedy features from Xt_orig1
Xt_orig1 = Xt_orig1[:, [0, 1, 3, 5, 7, 9, 12, 13, 14, 16, 19, 21, 22, 23, 24, 25, 26, 30, 31, 32, 33, 34, 35, 36]]
#Xt_orig1 = Xt_orig1[:,[0, 1, 3, 5, 8, 9, 11, 12, 13, 16, 18, 21, 22, 23, 25, 26, 27, 29, 30, 31, 32, 34, 35, 36, 37, 38]]
print "encoding original data..."
Xt_orig, keymap = utils.OneHotEncoder(Xt_orig1)
train_orig = Xt_orig[:num_train]
test_orig = Xt_orig[num_train:]
#print "use ACTION for data generation..."
y, X = data_io.load_data_pd('../data/train_orig.csv', use_labels=True)
_, X_test = data_io.load_data_pd('../data/test_orig.csv', use_labels=False)
oldTest = loadData('../data/test_orig.csv')
oldTrain= loadData('../data/train_orig.csv')
oldTrain = oldTrain[1:,1:]
oldTest = oldTest[1:,1:]
# # print "Grouping Data"
# # xd2 = utils.group_data2(oldTrain[:,:-1], degree=2) #skip last column
# # xd3 = utils.group_data2(oldTrain[:,:-1], degree=3) #skip last column
# # xtestd2 = utils.group_data2(oldTest[:,:-1], degree=2)
# # xtestd3 = utils.group_data2(oldTest[:,:-1], degree=3)
# # X_old = np.hstack((oldTrain, xd2, xd3))
def doCV():
SEED = 42
rnd = np.random.RandomState(SEED)
model_lr = linear_model.LogisticRegression(C=3)
model_rf = ensemble.RandomForestClassifier(
n_estimators=10, min_samples_split=10, compute_importances=False, n_jobs=2, random_state=rnd, verbose=2
)
print "loading data for random forest..."
y, X = data_io.load_data_pd("train_orig.csv", use_labels=True)
_, X_test = data_io.load_data_pd("test_orig.csv", use_labels=False)
xtrain = getRFX(X)
xtest = getRFX_test(X_test)
xtrain = xtrain[:, 1:]
xtest = xtest[:, 1:]
xtrain.dump("num_train.dat")
xtest.dump("num_test.dat")
print "dumped..!"
print "loading data for logistic regression..."
ysp, Xsp = data_io.load_data("train_orig.csv")
y_testsp, X_testsp = data_io.load_data("test_orig.csv", use_labels=False)
# === one-hot encoding === #
# we want to encode the category IDs encountered both in
# the training and the test set, so we fit the encoder on both
encoder = preprocessing.OneHotEncoder()
# print Xsp.shape, X_testsp.shape
encoder.fit(np.vstack((Xsp, X_testsp)))
Xsp = encoder.transform(Xsp) # Returns a sparse matrix (see numpy.sparse)
X_testsp = encoder.transform(X_testsp)
print "starting cross validation..."
nFeatures = X.shape[0]
niter = 10
cv = cross_validation.ShuffleSplit(nFeatures, n_iter=niter, test_size=0.2, random_state=rnd)
mean_auc = 0.0
i = 0
for train, test in cv:
xtrain = X.ix[train]
ytrain = y[train]
xtest = X.ix[test]
ytest = y[test]
xtrain_sp = Xsp[train]
xtest_sp = X_testsp[test]
ytrainsp = ysp[train]
xtrain = getRFX(xtrain)
xtest = getRFX_test(xtest)
xtrain = xtrain[:, 1:]
xtest = xtest[:, 1:]
print "fitting random forest...."
model_rf.fit(xtrain, ytrain)
preds_rf = model_rf.predict_proba(xtest)[:, 1]
print "fitting logistic regression..."
model_lr.fit(xtrain_sp, ytrainsp)
preds_lr = model_lr.predict_proba(xtest_sp)[:, 1]
preds = [np.mean(x) for x in zip(preds_rf, preds_lr)]
fpr, tpr, _ = metrics.roc_curve(ytest, preds)
roc_auc = metrics.auc(fpr, tpr)
print "AUC (fold %d/%d): %f" % (i + 1, niter, roc_auc)
mean_auc += roc_auc
i += 1
print "Mean AUC: ", mean_auc / niter
