Adding exponential decay logic

unit8co · dennisbader · Jun 17, 2024 · Dec 19, 2023 · Jan 4, 2024 · Jan 4, 2024
commit 5668c2ec4b2bf87fb0280df39f516a48588331b5
@@ -17,7 +17,7 @@
 from darts.logging import get_logger, raise_if, raise_if_not, raise_log
 from darts.timeseries import TimeSeries
 from darts.utils.utils import get_single_series, series2seq
-from darts.utils.timeseries_generation import constant_timeseries, linear_timeseries
+from darts.utils.timeseries_generation import constant_timeseries, linear_timeseries, exponential_timeseries
 
 logger = get_logger(__name__)
 
@@ -326,12 +326,19 @@ def create_lagged_data(
         times.append(times_i)
 
         if sample_weight:
+            weights = None
             if sample_weight == 'equal':
                 weights = constant_timeseries(1, start=times_i[0], end=times_i[-1], freq=times_i.freq).values()
-                sample_weights.append(weights)
             elif sample_weight == 'linear_decay':
                 weights = linear_timeseries(start=times_i[0], end=times_i[-1], freq=times_i.freq).values()
-                sample_weights.append(weights)
+            elif sample_weight == 'exponential_decay':
+                weights = exponential_timeseries(start=times_i[0], end=times_i[-1], freq=times_i.freq).values()
+            else:
+                raise ValueError(f"sample_weight {sample_weight} is not supported.")
+
+            sample_weights.append(weights)
+
+        # if instance of TimeSeries, convert to np.ndarray and append
 
     if concatenate:
         X = np.concatenate(X, axis=0)
@@ -596,7 +603,7 @@ def create_lagged_prediction_data(
         If the provided series do not share the same type of `time_index` (e.g. `target_series` uses a
         pd.RangeIndex, but `future_covariates` uses a `pd.DatetimeIndex`).
     """
-    X, _, times, _ = create_lagged_data(
+    X, _, times, _, _ = create_lagged_data(
         target_series=target_series,
         past_covariates=past_covariates,
         future_covariates=future_covariates,

diff --git a/darts/utils/timeseries_generation.py b/darts/utils/timeseries_generation.py
@@ -191,6 +191,64 @@ def linear_timeseries(
     )
 
 
+def exponential_timeseries(
+    start_value: float = 1,
+    end_value: float = 0,
+    start: Optional[Union[pd.Timestamp, int]] = pd.Timestamp("2000-01-01"),
+    end: Optional[Union[pd.Timestamp, int]] = None,
+    length: Optional[int] = None,
+    freq: Union[str, int] = None,
+    column_name: Optional[str] = "exponential",
+    dtype: np.dtype = np.float64,
+    decay_rate:int = 10,
+) -> TimeSeries:
+    """
+    Creates a univariate TimeSeries with a starting value of `start_value` that increases linearly such that
+    it takes on the value `end_value` at the last entry of the TimeSeries. This means that
+    the difference between two adjacent entries will be equal to
+    (`end_value` - `start_value`) / (`length` - 1).
+
+    Parameters
+    ----------
+    start_value
+        The value of the first entry in the TimeSeries.
+    end_value
+        The value of the last entry in the TimeSeries.
+    start
+        The start of the returned TimeSeries' index. If a pandas Timestamp is passed, the TimeSeries will have a pandas
+        DatetimeIndex. If an integer is passed, the TimeSeries will have a pandas RangeIndex index. Works only with
+        either `length` or `end`.
+    end
+        Optionally, the end of the returned index. Works only with either `start` or `length`. If `start` is
+        set, `end` must be of same type as `start`. Else, it can be either a pandas Timestamp or an integer.
+    length
+        Optionally, the length of the returned index. Works only with either `start` or `end`.
+    freq
+        The time difference between two adjacent entries in the returned index. In case `start` is a timestamp,
+        a DateOffset alias is expected; see
+        `docs <https://pandas.pydata.org/pandas-docs/stable/user_guide/TimeSeries.html#dateoffset-objects>`_.
+        By default, "D" (daily) is used.
+        If `start` is an integer, `freq` will be interpreted as the step size in the underlying RangeIndex.
+        The freq is optional for generating an integer index (if not specified, 1 is used).
+    column_name
+        Optionally, the name of the value column for the returned TimeSeries
+    dtype
+        The desired NumPy dtype (np.float32 or np.float64) for the resulting series
+
+    Returns
+    -------
+    TimeSeries
+        A exponentially decreasing TimeSeries created as indicated above.
+    """
+
+    index = generate_index(start=start, end=end, freq=freq, length=length)
+    time_steps = np.linspace(start_value, end_value, len(index), dtype=dtype)
+    values = np.exp(-decay_rate * (1 - time_steps))
+    return TimeSeries.from_times_and_values(
+        index, values, freq=freq, columns=pd.Index([column_name])
+    )
+
+
 def sine_timeseries(
     value_frequency: float = 0.1,
     value_amplitude: float = 1.0,