def train_and_predict(self, X_train, X_valid, y_train, y_valid, X_test, catboost_params):
# データセットを生成する
train_pool = Pool(X_train, label=y_train)
valid_pool = Pool(X_valid, label=y_valid)
evals = [valid_pool]
model = CatBoost(catboost_params)
# ロガーの作成
with timer("Training"):
# 上記のパラメータでモデルを学習する
model.fit(
train_pool,
cat_features= data.get_category_columns(X_train),
eval_set=evals,
# 50 ラウンド経過しても性能が向上しないときは学習を打ち切る
early_stopping_rounds=50,
)
y_train_pred = model.predict(train_pool).reshape(-1,1)
#検証データの予測
y_valid_pred = model.predict(valid_pool).reshape(-1,1)
# テストデータの予測
y_pred = model.predict(X_test).reshape(-1,1)
return y_train_pred, y_valid_pred, y_pred, model
def train_and_predict(self, X_train, X_valid, y_train, y_valid, X_test,
xgbm_params):
# データセットを生成する
dtrain = xgb.DMatrix(X_train, label=y_train)
dtest = xgb.DMatrix(X_valid, label=y_valid)
evals = [(dtrain, 'train'), (dtest, 'eval')]
evals_result = {}
# ロガーの作成
logger = logging.getLogger('main')
callbacks = [log_evaluation_xgbm(logger, period=50)]
with timer("Training"):
# 上記のパラメータでモデルを学習する
model = xgb.train(
xgbm_params,
dtrain,
evals=evals,
evals_result=evals_result,
num_boost_round=1000,
# 30 ラウンド経過しても性能が向上しないときは学習を打ち切る
early_stopping_rounds=50,
# ログ
callbacks=callbacks)
y_train_pred = model.predict(xgb.DMatrix(X_train)).reshape(-1, 1)
#検証データの予測
y_valid_pred = model.predict(xgb.DMatrix(X_valid)).reshape(-1, 1)
# テストデータの予測
y_pred = model.predict(xgb.DMatrix(X_test)).reshape(-1, 1)
return y_train_pred, y_valid_pred, y_pred, model
def run(self):
with timer(self.name):
self.create_features()
prefix = self.prefix + '_' if self.prefix else ''
suffix = '_' + self.suffix if self.suffix else ''
self.train.columns = prefix + self.train.columns + suffix
self.test.columns = prefix + self.test.columns + suffix
return self
def run(self):
with timer(self.name):
self.create_features()
self.historical_transactions = reduce_mem_usage(
self.historical_transactions)
self.new_merchant_transactions = reduce_mem_usage(
self.new_merchant_transactions)
prefix = self.prefix + '_' if self.prefix else ''
suffix = '_' + self.suffix if self.suffix else ''
self.historical_transactions.columns = prefix + self.historical_transactions.columns + suffix
self.new_merchant_transactions.columns = prefix + self.new_merchant_transactions.columns + suffix
return self
def train_and_predict(self, X_train, X_valid, y_train, y_valid, X_test, lasso_params):
# ロガーの作成
logger = logging.getLogger('main')
#標準化
sc = StandardScaler()
X_train_std = sc.fit_transform(X_train.values)
X_valid_std = sc.fit_transform(X_valid.values)
X_test_std = sc.fit_transform(X_test.values)
with timer("Training"):
model = Lasso(**lasso_params)
model.fit(X_train_std, y_train)
y_train_pred = model.predict(X_train_std).reshape(-1,1)
#検証データの予測
y_valid_pred = model.predict(X_valid_std).reshape(-1,1)
#テストデータの予測
y_test_pred = model.predict(X_test_std).reshape(-1,1)
return y_train_pred, y_valid_pred, y_test_pred, model
def train_and_predict(self, X_train, X_valid, y_train, y_valid, X_test,
lgbm_params):
# データセットを生成する
lgb_train = lgb.Dataset(X_train, y_train)
lgb_eval = lgb.Dataset(X_valid, y_valid, reference=lgb_train)
evals_result = {}
# ロガーの作成
logger = logging.getLogger('main')
callbacks = [log_evaluation_lgbm(logger, period=50)]
with timer("Training"):
# 上記のパラメータでモデルを学習する
model = lgb.train(
lgbm_params,
lgb_train,
# モデルの評価用データを渡す
valid_sets=lgb_eval,
evals_result=evals_result,
# 最大で 1000 ラウンドまで学習する
num_boost_round=1000,
# 30 ラウンド経過しても性能が向上しないときは学習を打ち切る
early_stopping_rounds=50,
# ログ
callbacks=callbacks)
y_train_pred = model.predict(
X_train, num_iteration=model.best_iteration).reshape(-1, 1)
#検証データの予測
y_valid_pred = model.predict(
X_valid, num_iteration=model.best_iteration).reshape(-1, 1)
# テストデータの予測
y_pred = model.predict(X_test,
num_iteration=model.best_iteration).reshape(
-1, 1)
return y_train_pred, y_valid_pred, y_pred, model
