def main():
train, test, macro = data_utils.load_data()
train.fillna(0, inplace=True)
test.fillna(0)
mult = .969
train['price_doc'] = train["price_doc"] * mult + 10
# train['price_doc'] = np.log1p(train['price_doc'])
y_train = train['price_doc']
id_train = train['id']
train.drop(['id', 'price_doc'], axis=1, inplace=True)
id_test = test['id']
test.drop(['id'], axis=1, inplace=True)
# 合并训练集和测试集
conbined_data = pd.concat([train[test.columns.values], test])
conbined_data.drop(['timestamp'], axis=1, inplace=True)
print "conbined_data:", conbined_data.shape
# Deal with categorical values
for c in conbined_data.columns:
if conbined_data[c].dtype == 'object':
lbl = preprocessing.LabelEncoder()
lbl.fit(list(conbined_data[c].values))
conbined_data[c] = lbl.transform(list(conbined_data[c].values))
del conbined_data['school_education_centers_raion_ratio_dis']
del conbined_data['preschool_education_centers_raion_ratio_dis']
del conbined_data['sport_objects_raion_ratio_dis']
del conbined_data['additional_education_raion_ratio_dis']
del conbined_data['0_6_all_vs_preschool_quota_dis']
scaler = StandardScaler()
conbined_data = scaler.fit_transform(conbined_data)
train = conbined_data[:train.shape[0], :]
test = conbined_data[train.shape[0]:, :]
test_size = (1.0 * test.shape[0]) / train.shape[0]
print "submit test size:", test_size
ntrain = train.shape[0]
ntest = test.shape[0]
n_folds = 5
random_seed = 0
kfold = KFold(n_splits=n_folds, shuffle=True, random_state=random_seed)
oof_train = np.zeros((ntrain,))
oof_test = np.zeros((ntest,))
oof_test_skf = np.empty((n_folds, ntest))
for i, (train_index, test_index) in enumerate(kfold.split(train)):
print 'fold-{}: train: {}, test: {}'.format(i, train_index, test_index)
x_tr = train[train_index]
y_tr = y_train[train_index]
x_te = train[test_index]
alphas = np.array([1,0.5, 0.1, 0.05,0.01, 0.005])
solverOptions = (['svd', 'cholesky', 'sparse_cg', 'sag'])
# create and fit a ridge regression model, testing each alpha
model = Ridge(normalize=True, fit_intercept=True) #We have chosen to just normalize the data by default, you could GridsearchCV this is you wanted
grid = GridSearchCV(estimator=model, param_grid=dict(alpha=alphas, solver=solverOptions))
grid.fit(x_tr, y_tr)
# summarize the results of the grid search
print 'best_score', grid.best_score_
print 'alphas:', grid.best_estimator_.alpha
print 'solverOptions:', grid.best_estimator_.solver
model = Ridge(normalize=True, alpha=grid.best_estimator_.alpha, fit_intercept=True,
solver=grid.best_estimator_.solver) # paramters tuned using GridSearchCV
model.fit(x_tr, y_tr)
train_rmse = mean_squared_error(y_tr, model.predict(x_tr))
print 'train_rmse =', np.sqrt(train_rmse)
oof_train[test_index] = model.predict(x_te)
oof_test_skf[i, :] = model.predict(test)
oof_test[:] = oof_test_skf.mean(axis=0)
# 保存 oof 结果
train_predict = pd.DataFrame({'id': id_train,
'ridge_oof_predict': oof_train})
test_predict = pd.DataFrame({'id': id_test,
'ridge_oof_predict': oof_test})
train_predict.to_csv(Configure.train_cv_result_for_model_stacking.format('ridge', time.strftime('%Y-%m-%d_%H:%M:%S', time.localtime(time.time()))),
index=False)
test_predict.to_csv(Configure.test_cv_result_for_model_stacking.format('ridge', time.strftime('%Y-%m-%d_%H:%M:%S', time.localtime(time.time()))),
index=False)
sys.path.append(module_path)
import numpy as np
import pandas as pd
# remove warnings
import warnings
warnings.filterwarnings('ignore')
import xgboost as xgb
from sklearn.metrics import mean_squared_error
# my own module
from features import data_utils
from sklearn import preprocessing
from conf.configure import Configure
train, test, macro = data_utils.load_data()
print 'train:', train.shape
print 'test:', test.shape
# Deal with categorical values
for c in train.columns:
if train[c].dtype == 'object':
lbl = preprocessing.LabelEncoder()
lbl.fit(list(train[c].values))
train[c] = lbl.transform(list(train[c].values))
for c in test.columns:
if test[c].dtype == 'object':
lbl = preprocessing.LabelEncoder()
lbl.fit(list(test[c].values))
test[c] = lbl.transform(list(test[c].values))
def main():
train, test, macro = data_utils.load_data()
ylog_train_all = train['price_doc']
train.drop(['id', 'price_doc'], axis=1, inplace=True)
test.drop(['id'], axis=1, inplace=True)
# 合并训练集和测试集
conbined_data = pd.concat([train[test.columns.values], test])
conbined_data.columns = test.columns.values
str_columns = conbined_data.select_dtypes(include=['object']).columns.values.tolist()
# dummy code
dummies_data = pd.get_dummies(conbined_data[str_columns])
conbined_data[dummies_data.columns] = dummies_data[dummies_data.columns]
conbined_data.drop(str_columns, axis=1, inplace=True)
train = conbined_data.iloc[:train.shape[0], :]
test = conbined_data.iloc[train.shape[0]:, :]
test_size = (1.0 * test.shape[0]) / (train.shape[0] + test.shape[0])
print "submit test size:", test_size
# Convert to numpy values
X_all = train.values
print(X_all.shape)
# Create a validation set, with last 20% of data
num_train = train.shape[0]
num_val = int(num_train * 0.2)
X_train_all = X_all[:num_train]
X_train = X_all[:num_train - num_val]
X_val = X_all[num_train - num_val:num_train]
ylog_train = ylog_train_all[:-num_val]
ylog_val = ylog_train_all[-num_val:]
X_test = test
df_columns = train.columns
print('X_train_all shape is', X_train_all.shape)
print('X_train shape is', X_train.shape)
print('y_train shape is', ylog_train.shape)
print('X_val shape is', X_val.shape)
print('y_val shape is', ylog_val.shape)
print('X_test shape is', X_test.shape)
dtrain = xgb.DMatrix(X_train, ylog_train, feature_names=df_columns)
dval = xgb.DMatrix(X_val, ylog_val, feature_names=df_columns)
xgb_params = {
'eta': 0.01,
'max_depth': 5,
'subsample': 0.7,
'booster': 'dart',
'colsample_bytree': 0.7,
'objective': 'reg:linear',
'eval_metric': 'rmse',
'silent': 1,
'seed': 100
}
num_round = 500
xgb_params['nthread'] = 24
# param['eval_metric'] = "auc"
plst = xgb_params.items()
plst += [('eval_metric', 'rmse')]
evallist = [(dval, 'eval')]
bst = xgb.train(plst, dtrain, num_round, evallist, early_stopping_rounds=20, verbose_eval=10)
num_boost_round = bst.best_iteration + 1
print 'best_iteration: ', num_boost_round
def main():
train, test, macro = data_utils.load_data()
mult = .969
train['price_doc'] = train["price_doc"] * mult + 10
# train['price_doc'] = np.log1p(train['price_doc'])
ylog_train_all = train['price_doc']
id_train = train['id']
train.drop(['id', 'price_doc'], axis=1, inplace=True)
submit_ids = test['id']
test.drop(['id'], axis=1, inplace=True)
# 合并训练集和测试集
conbined_data = pd.concat([train[test.columns.values], test])
# macro_cols = ["balance_trade", "balance_trade_growth", "eurrub", "average_provision_of_build_contract",
# "micex_rgbi_tr", "micex_cbi_tr", "deposits_rate", "mortgage_value", "mortgage_rate",
# "income_per_cap", "rent_price_4+room_bus", "museum_visitis_per_100_cap", "apartment_build", "timestamp"]
# conbined_data = pd.merge_ordered(conbined_data, macro[macro_cols], on='timestamp', how='left')
conbined_data.drop(['timestamp'], axis=1, inplace=True)
print "conbined_data:", conbined_data.shape
# Deal with categorical values
for c in conbined_data.columns:
if conbined_data[c].dtype == 'object':
lbl = preprocessing.LabelEncoder()
lbl.fit(list(conbined_data[c].values))
conbined_data[c] = lbl.transform(list(conbined_data[c].values))
train = conbined_data.iloc[:train.shape[0], :]
test = conbined_data.iloc[train.shape[0]:, :]
test_size = (1.0 * test.shape[0]) / train.shape[0]
print "submit test size:", test_size
# Convert to numpy values
X_all = train.values
# Create a validation set, with last 20% of data
num_train = int(train.shape[0] / (1 + test_size))
X_train_all = X_all
X_train = X_all[:num_train]
X_val = X_all[num_train:]
ylog_train = ylog_train_all[:num_train]
ylog_val = ylog_train_all[num_train:]
X_test = test
print "validate size:", 1.0 * X_val.shape[0] / X_train.shape[0]
df_columns = train.columns
print('X_train_all shape is', X_train_all.shape)
print('X_train shape is', X_train.shape)
print('y_train shape is', ylog_train.shape)
print('X_val shape is', X_val.shape)
print('y_val shape is', ylog_val.shape)
print('X_test shape is', X_test.shape)
dtrain_all = xgb.DMatrix(X_train_all,
ylog_train_all,
feature_names=df_columns)
dtrain = xgb.DMatrix(X_train, ylog_train, feature_names=df_columns)
dval = xgb.DMatrix(X_val, ylog_val, feature_names=df_columns)
dtest = xgb.DMatrix(X_test, feature_names=df_columns)
xgb_params = {
'eta': 0.05,
'max_depth': 5,
'subsample': 0.7,
'colsample_bytree': 0.7,
'objective': 'reg:linear',
'eval_metric': 'rmse',
'silent': 1
}
num_round = 1000
xgb_params['nthread'] = 24
evallist = [(dval, 'eval')]
bst = xgb.train(xgb_params,
dtrain,
num_round,
evallist,
early_stopping_rounds=40,
verbose_eval=10)
train_rmse = mean_squared_error(ylog_train, bst.predict(dtrain))
val_rmse = mean_squared_error(ylog_val, bst.predict(dval))
print 'train_rmse =', np.sqrt(train_rmse), ', val_rmse =', np.sqrt(
val_rmse)
num_boost_round = bst.best_iteration
print 'best_iteration: ', num_boost_round
model = xgb.train(dict(xgb_params, silent=1),
dtrain_all,
num_boost_round=num_boost_round)
print 'predict submit...'
y_pred = model.predict(dtest)
# y_pred = np.exp(ylog_pred) - 1
df_sub = pd.DataFrame({'id': submit_ids, 'price_doc': y_pred})
df_sub.to_csv('xgboost_model_4.csv', index=False) # 0.31499
# save model
model.save_model('xgboost_model4.model')
def main():
train, test, macro = data_utils.load_data()
mult = .969
train['price_doc'] = train["price_doc"] * mult + 10
y_train = train['price_doc']
id_train = train['id']
train.drop(['id', 'price_doc'], axis=1, inplace=True)
id_test = test['id']
test.drop(['id'], axis=1, inplace=True)
# 合并训练集和测试集
conbined_data = pd.concat([train[test.columns.values], test])
conbined_data.drop(['timestamp'], axis=1, inplace=True)
print "conbined_data:", conbined_data.shape
# Deal with categorical values
for c in conbined_data.columns:
if conbined_data[c].dtype == 'object':
lbl = preprocessing.LabelEncoder()
lbl.fit(list(conbined_data[c].values))
conbined_data[c] = lbl.transform(list(conbined_data[c].values))
train = conbined_data.iloc[:train.shape[0], :]
test = conbined_data.iloc[train.shape[0]:, :]
test_size = (1.0 * test.shape[0]) / train.shape[0]
print "submit test size:", test_size
# Convert to numpy values
train = train.values
test = test.values
ntrain = train.shape[0]
ntest = test.shape[0]
n_folds = 5
random_seed = 0
kfold = KFold(n_splits=n_folds, shuffle=True, random_state=random_seed)
oof_train = np.zeros((ntrain, ))
oof_test = np.zeros((ntest, ))
oof_test_skf = np.empty((n_folds, ntest))
xgb_params = {
'eta': 0.05,
'max_depth': 5,
'subsample': 0.7,
'colsample_bytree': 0.7,
'objective': 'reg:linear',
'eval_metric': 'rmse',
'silent': 1,
'nthread': 24
}
num_round = 1000
dtest = xgb.DMatrix(test)
for i, (train_index, test_index) in enumerate(kfold.split(train)):
print 'fold-{}: train: {}, test: {}'.format(i, train_index, test_index)
x_tr = train[train_index]
y_tr = y_train[train_index]
x_te = train[test_index]
dtrain = xgb.DMatrix(x_tr, y_tr)
dx_te = xgb.DMatrix(x_te)
cv_output = xgb.cv(dict(xgb_params, silent=1),
dtrain,
num_boost_round=num_round,
early_stopping_rounds=40)
num_boost_round = len(cv_output)
print 'best_iteration: ', num_boost_round
model = xgb.train(dict(xgb_params, silent=1),
dtrain,
num_boost_round=num_boost_round)
train_rmse = mean_squared_error(y_tr, model.predict(dtrain))
print 'train_rmse =', np.sqrt(train_rmse)
oof_train[test_index] = model.predict(dx_te)
oof_test_skf[i, :] = model.predict(dtest)
oof_test[:] = oof_test_skf.mean(axis=0)
# 保存 oof 结果
train_predict = pd.DataFrame({
'id': id_train,
'xgboost_oof_predict': oof_train
})
test_predict = pd.DataFrame({
'id': id_test,
'xgboost_oof_predict': oof_test
})
train_predict.to_csv(Configure.train_cv_result_for_model_stacking.format(
'xgboost',
time.strftime('%Y-%m-%d_%H:%M:%S', time.localtime(time.time()))),
index=False)
test_predict.to_csv(Configure.test_cv_result_for_model_stacking.format(
'xgboost',
time.strftime('%Y-%m-%d_%H:%M:%S', time.localtime(time.time()))),
index=False)