def test_minute_frequency():
prediction_length = 1
timeseries = {
TIMESERIES_KEYS.START: "2021-01-15 12:40:00",
TIMESERIES_KEYS.TARGET: np.array([12, 13]),
TIMESERIES_KEYS.TARGET_NAME: "target",
TIMESERIES_KEYS.TIME_COLUMN_NAME: "date",
}
frequency = "20min"
gluon_dataset = ListDataset([timeseries], freq=frequency)
model = Model(
"simplefeedforward",
model_parameters={"activated": True, "kwargs": {}},
frequency=frequency,
prediction_length=prediction_length,
epoch=1,
batch_size=8,
num_batches_per_epoch=5,
)
evaluation_forecasts_df = model.train_evaluate(gluon_dataset, gluon_dataset, make_forecasts=True, retrain=True)[2]
assert evaluation_forecasts_df["index"].iloc[0] == pd.Timestamp("2021-01-15 13:00:00")
trained_model = TrainedModel(
predictor=model.predictor,
gluon_dataset=gluon_dataset,
prediction_length=prediction_length,
quantiles=[0.5],
include_history=True,
)
trained_model.predict()
forecasts_df = trained_model.get_forecasts_df(session="2021-01-01", model_label="TEST")
assert forecasts_df["date"].iloc[0] == pd.Timestamp("2021-01-15 13:20:00")
def setup_method(self):
self.session_name = datetime.utcnow().isoformat() + "Z"
self.model_label = "TrivialIdentity"
self.trained_model = TrainedModel(
predictor=self.predictor,
gluon_dataset=self.gluon_dataset,
prediction_length=self.prediction_length,
quantiles=[0.1, 0.5, 0.9],
include_history=True,
)
def test_simplefeedforward_external_features_predictions(self):
external_features_future_df = pd.DataFrame(
{
"date": ["2018-01-10", "2018-01-11", "2018-01-13", "2018-01-14"],
"key": [1, 1, 2, 2],
"ext_feat": [0, 0, 0, 1],
}
)
external_features_future_df["date"] = pd.to_datetime(external_features_future_df["date"]).dt.tz_localize(tz=None)
gluon_dataset = add_future_external_features(self.test_list_dataset, external_features_future_df, self.prediction_length, self.frequency)
trained_model = TrainedModel(
model_name=self.model_name,
predictor=self.model.predictor,
gluon_dataset=gluon_dataset,
prediction_length=self.prediction_length,
quantiles=[0.1, 0.5, 0.9],
include_history=True,
)
trained_model.predict()
forecasts_df = trained_model.get_forecasts_df(session="1234")
assert len(forecasts_df.index) == 13
gluon_train_dataset = model_selection.get_gluon_train_dataset()
external_features = external_features_check(
gluon_train_dataset, params["external_features_future_dataset"])
if external_features:
external_features_future_df = params[
"external_features_future_dataset"].get_dataframe()
gluon_train_dataset = add_future_external_features(
gluon_train_dataset, external_features_future_df,
predictor.prediction_length, predictor.freq)
trained_model = TrainedModel(
predictor=predictor,
gluon_dataset=gluon_train_dataset,
prediction_length=params["prediction_length"],
quantiles=params["quantiles"],
include_history=params["include_history"],
history_length_limit=params["history_length_limit"],
)
trained_model.predict()
logger.info(
"Forecasting future values: Done in {:.2f} seconds".format(perf_counter() -
start))
forecasts_df = trained_model.get_forecasts_df(
session=model_selection.get_session_name(),
model_label=model_selection.get_model_label())
params["output_dataset"].write_with_schema(forecasts_df)
class TestTrainedModel:
def setup_class(self):
self.frequency = "D"
self.prediction_length = 2
timeseries_0 = {
TIMESERIES_KEYS.START:
"2018-01-01",
TIMESERIES_KEYS.TARGET:
np.array([12, 13, 14, 15]),
TIMESERIES_KEYS.TARGET_NAME:
"sales",
TIMESERIES_KEYS.TIME_COLUMN_NAME:
"date",
TIMESERIES_KEYS.FEAT_DYNAMIC_REAL:
np.array([[1, 0, 0, 0, 0, 1], [0, 0, 0, 1, 1, 0]]),
TIMESERIES_KEYS.FEAT_DYNAMIC_REAL_COLUMNS_NAMES:
["is_holiday", "is_weekend"],
TIMESERIES_KEYS.IDENTIFIERS: {
"store": 1,
"item": 1
},
}
timeseries_1 = {
TIMESERIES_KEYS.START:
"2018-01-01",
TIMESERIES_KEYS.TARGET:
np.array([2, 3, 4, 5]),
TIMESERIES_KEYS.TARGET_NAME:
"sales",
TIMESERIES_KEYS.TIME_COLUMN_NAME:
"date",
TIMESERIES_KEYS.FEAT_DYNAMIC_REAL:
np.array([[1, 0, 0, 0, 0, 1], [0, 0, 0, 1, 1, 0]]),
TIMESERIES_KEYS.FEAT_DYNAMIC_REAL_COLUMNS_NAMES:
["is_holiday", "is_weekend"],
TIMESERIES_KEYS.IDENTIFIERS: {
"store": 1,
"item": 2
},
}
self.gluon_dataset = ListDataset([timeseries_0, timeseries_1],
freq=self.frequency)
self.predictor = IdentityPredictor(prediction_length=2,
freq="D",
num_samples=100)
def setup_method(self):
self.session_name = datetime.utcnow().isoformat() + "Z"
self.model_label = "TrivialIdentity"
self.trained_model = TrainedModel(
model_name="trivial_identity",
predictor=self.predictor,
gluon_dataset=self.gluon_dataset,
prediction_length=self.prediction_length,
quantiles=[0.1, 0.5, 0.9],
include_history=True,
)
def test_predict(self):
self.trained_model.predict()
forecasts_df = self.trained_model.get_forecasts_df(
session=self.session_name)
expected_forecasts_columns = [
"date",
"store",
"item",
"is_holiday",
"is_weekend",
"sales",
"forecast_lower_sales",
"forecast_sales",
"forecast_upper_sales",
METRICS_DATASET.SESSION,
METRICS_DATASET.MODEL_COLUMN,
ROW_ORIGIN.COLUMN_NAME,
]
assert set(expected_forecasts_columns) == set(forecasts_df.columns)
future_df = forecasts_df[forecasts_df["item"] == 2].iloc[0:2]
history_df = forecasts_df[forecasts_df["item"] == 2].iloc[2:]
assert future_df["sales"].count() == 0 and history_df["sales"].count(
) == 4
assert future_df["forecast_sales"].count(
) == 2 and history_df["forecast_sales"].count() == 0