Initial commit

evalml/pipelines/multiseries_regression_pipeline.py

@@ -1,4 +1,5 @@
 """Pipeline base class for time series regression problems."""
+import pandas as pd
 from woodwork.statistics_utils import infer_frequency
 
 from evalml.pipelines.time_series_regression_pipeline import (
@@ -137,3 +138,39 @@ def predict_in_sample(
         # Index will start at the unstacked index, so we need to reset it to the original index
         stacked_predictions.index = X.index
         return stacked_predictions
+
+    def get_forecast_period(self, X):
+        """Generates all possible forecasting time points based on latest data point in X.
+
+        For the multiseries case, each time stamp is duplicated for each unique value in `X`'s `series_id` column.
+
+        Args:
+            X (pd.DataFrame, np.ndarray): Data the pipeline was trained on of shape [n_samples_train, n_feautures].
+
+        Raises:
+            ValueError: If pipeline is not trained.
+
+        Returns:
+            pd.DataFrame: Dataframe containing a column with datetime periods from `gap` to `forecast_horizon + gap`
+            per unique `series_id` value.
+        """
+        dates = super().get_forecast_period(X)
+        dates.name = self.time_index
+        series_id_values = X[self.series_id].unique()
+
+        new_period_df = dates.to_frame().merge(
+            pd.Series(series_id_values, name=self.series_id),
+            how="cross",
+        )
+
+        # Generate new numeric index
+        num_idx = pd.Series(
+            range(
+                dates.index[0] + (self.gap * len(series_id_values)),
+                dates.index[0]
+                + (self.gap * len(series_id_values))
+                + len(new_period_df),
+            ),
+        )
+        new_period_df.index = num_idx
+        return new_period_df

evalml/tests/pipeline_tests/regression_pipeline_tests/test_multiseries_regression_pipeline.py

@@ -1,5 +1,7 @@
+import numpy as np
 import pandas as pd
 import pytest
+from pandas._testing import assert_series_equal
 
 from evalml.pipelines import MultiseriesRegressionPipeline
 from evalml.preprocessing import split_multiseries_data
@@ -157,3 +159,83 @@ def test_multiseries_pipeline_predict(
             dtype="float64",
         )
     pd.testing.assert_series_equal(y_pred, expected)
+
+
+@pytest.mark.parametrize("forecast_horizon,gap", [[3, 0], [5, 2], [2, 5]])
+@pytest.mark.parametrize("numeric_idx", [True, False])
+def test_time_series_get_forecast_period(
+    forecast_horizon,
+    gap,
+    numeric_idx,
+    multiseries_ts_data_stacked,
+    component_graph,
+    pipeline_parameters,
+):
+    X, y = multiseries_ts_data_stacked
+    if numeric_idx:
+        X = X.reset_index(drop=True)
+
+    pipeline_parameters["pipeline"]["forecast_horizon"] = forecast_horizon
+    pipeline_parameters["Time Series Featurizer"]["forecast_horizon"] = forecast_horizon
+    pipeline_parameters["Baseline Multiseries"]["forecast_horizon"] = forecast_horizon
+    pipeline_parameters["pipeline"]["gap"] = gap
+    pipeline_parameters["Time Series Featurizer"]["gap"] = gap
+    pipeline_parameters["Baseline Multiseries"]["gap"] = gap
+
+    clf = MultiseriesRegressionPipeline(component_graph, pipeline_parameters)
+
+    with pytest.raises(
+        ValueError,
+        match="Pipeline must be fitted before getting forecast.",
+    ):
+        clf.get_forecast_period(X)
+
+    clf.fit(X, y)
+    result = clf.get_forecast_period(X)
+
+    len_unique_series_id = len(X["series_id"].unique())
+
+    assert result.shape[0] == forecast_horizon * len_unique_series_id
+    assert all(
+        result.index
+        == range(
+            len(X) + (gap * len_unique_series_id),
+            len(X)
+            + (gap * len_unique_series_id)
+            + (forecast_horizon * len_unique_series_id),
+        ),
+    )
+    assert result.iloc[0]["date"] == X.iloc[-1]["date"] + np.timedelta64(
+        1 + gap,
+        clf.frequency,
+    )
+    assert np.issubdtype(result.dtypes["date"], np.datetime64)
+    assert list(result.columns) == ["date", "series_id"]
+
+
+@pytest.mark.parametrize("forecast_horizon,gap", [[3, 0], [5, 2], [2, 5]])
+def test_time_series_get_forecast_predictions(
+    forecast_horizon,
+    gap,
+    multiseries_ts_data_stacked,
+    component_graph,
+    pipeline_parameters,
+):
+    X, y = multiseries_ts_data_stacked
+
+    X_train, y_train = X.iloc[:25], y.iloc[:25]
+    X_validation = X.iloc[25 + (gap * 5) : 25 + (gap * 5) + (forecast_horizon * 5)]
+
+    pipeline_parameters["pipeline"]["forecast_horizon"] = forecast_horizon
+    pipeline_parameters["Time Series Featurizer"]["forecast_horizon"] = forecast_horizon
+    pipeline_parameters["Baseline Multiseries"]["forecast_horizon"] = forecast_horizon
+    pipeline_parameters["pipeline"]["gap"] = gap
+    pipeline_parameters["Time Series Featurizer"]["gap"] = gap
+    pipeline_parameters["Baseline Multiseries"]["gap"] = gap
+
+    clf = MultiseriesRegressionPipeline(component_graph, pipeline_parameters)
+
+    clf.fit(X_train, y_train)
+    forecast_preds = clf.get_forecast_predictions(X=X_train, y=y_train)
+    X_val_preds = clf.predict(X_validation, X_train=X_train, y_train=y_train)
+    assert_series_equal(forecast_preds, X_val_preds)