def run(self):
config = GlobalParams().config
store_id = self.store_id
logger.info(f"Training model for store_id: {store_id}")
sales = self.input().load()
train_dates = sorted(sales.date.unique())
valid_start_date = min(train_dates[-config["valid_num_days"]:])
X_train = sales[sales.date < valid_start_date].drop(
[config["pred_target_col"], config["pred_date_col"], "date"],
axis=1)
y_train = sales[sales.date < valid_start_date][
config["pred_target_col"]]
X_valid = sales[sales.date >= valid_start_date].drop(
[config["pred_target_col"], config["pred_date_col"], "date"],
axis=1)
y_valid = sales[sales.date >= valid_start_date][
config["pred_target_col"]]
model = LGBMRegressor(**GlobalParams().config["lgb_params"])
model.fit(
X_train,
y_train,
eval_set=[(X_train, y_train), (X_valid, y_valid)],
eval_metric="rmse",
early_stopping_rounds=10,
)
self.output().save(model)
def run(self):
config = GlobalParams().config
test_start_date = config["test_start_date"]
test_end_date = config["test_end_date"]
logger.info("Preparing test data for date range: {} - {}".format(
test_start_date, test_end_date))
sales_base = self.input()["sales_base"].load()
sales = sales_base[(test_start_date <= sales_base.date)
& (sales_base.date <= test_end_date)]
del sales_base
sales_features = self.input()["sales_ts"].load()
sales_features = sales_features[
(test_start_date <= sales_features.date)
& (sales_features.date <= test_end_date)]
columns = ["id", "date"
] + list(set(sales_features.columns) - set(sales.columns))
sales = sales.merge(
sales_features[[c for c in sales_features.columns
if c in columns]],
how="left",
on=["id", "date"],
)
del sales_features
self.output().save(sales)
def run(self):
config = GlobalParams().config
train_start_date = config["train_start_date"]
train_end_date = config["train_end_date"]
logger.info("Preparing train data for range: {} - {}".format(
train_start_date, train_end_date))
sales_base = self.input()["sales_base"].load()
sales = sales_base[(train_start_date <= sales_base.date)
& (sales_base.date <= train_end_date)]
del sales_base
sales_features = self.input()["sales_ts"].load()
sales_features = sales_features[
(train_start_date <= sales_features.date)
& (sales_features.date <= train_end_date)]
columns = ["id", "date"
] + list(set(sales_features.columns) - set(sales.columns))
sales = sales.merge(
sales_features[[c for c in sales_features.columns
if c in columns]],
how="left",
on=["id", "date"],
)
del sales_features
shift_days = self.pred_week * NUM_DAYS_1_WEEK
sales[config["pred_target_col"]] = sales.groupby("id")[
config["target_col"]].shift(-shift_days)
sales[config["pred_date_col"]] = sales.groupby("id")["date"].shift(
-shift_days)
sales = sales[sales[config["pred_date_col"]].notnull()]
self.output().save(sales)
def run(self):
config = GlobalParams().config
preds = []
for store_id in config["store_list"]:
for pred_week in config["pred_week_list"]:
preds.append(
RunPredictionStoreWeek(
store_id=store_id,
pred_week=pred_week).output().load())
preds_all = pd.concat(preds)
preds_all = preds_all.sort_values(["id", config["pred_date_col"]],
ignore_index=True)
self.output().save(preds_all)
def run(self):
config = GlobalParams().config
model = self.input()["model"].load()
X_test = self.input()["test_data"].load()
preds = X_test[["id", "date"]].copy()
preds[config["pred_date_col"]] = preds["date"] + np.timedelta64(
self.pred_week * NUM_DAYS_1_WEEK, "D")
X_test = X_test.drop("date", axis=1)
preds[config["pred_target_col"]] = model.predict(X_test)
self.output().save(preds)
# Clean-up files that are not needed anymore
file = (PrepareTrainData(store_id=self.store_id,
pred_week=self.pred_week).output().path)
if os.path.exists(file):
os.remove(file)
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
GlobalParams(config_file=self.config_file)
def requires(self):
return {
s: RunPredictionStore(store_id=s)
for s in GlobalParams().config["store_list"]
}
def requires(self):
return {
w: RunPredictionStoreWeek(store_id=self.store_id, pred_week=w)
for w in GlobalParams().config["pred_week_list"]
}