"""Prophet is a procedure for forecasting time series data based on an additive model where non-linear trends are fit with yearly, weekly, and daily seasonality, plus holiday effects. It works best with time series that have strong seasonal effects and several seasons of historical data. Prophet is robust to missing data and shifts in the trend, and typically handles outliers well."""
import copy
from typing import Dict, Hashable, List, Optional, Union
import numpy as np
import pandas as pd
from skopt.space import Real
from evalml.model_family import ModelFamily
from evalml.pipelines.components.estimators import Estimator
from evalml.problem_types import ProblemTypes
from evalml.utils import import_or_raise, infer_feature_types
from evalml.utils.gen_utils import classproperty
[docs]class ProphetRegressor(Estimator):
"""Prophet is a procedure for forecasting time series data based on an additive model where non-linear trends are fit with yearly, weekly, and daily seasonality, plus holiday effects. It works best with time series that have strong seasonal effects and several seasons of historical data. Prophet is robust to missing data and shifts in the trend, and typically handles outliers well.
More information here: https://facebook.github.io/prophet/
Args:
time_index (str): Specifies the name of the column in X that provides the datetime objects. Defaults to None.
changepoint_prior_scale (float): Determines the strength of the sparse prior for fitting on rate changes. Increasing
this value increases the flexibility of the trend. Defaults to 0.05.
seasonality_prior_scale (int): Similar to changepoint_prior_scale. Adjusts the extent to which the seasonality model will fit the data.
Defaults to 10.
holidays_prior_scale (int): Similar to changepoint_prior_scale. Adjusts the extent to which holidays will fit the data.
Defaults to 10.
seasonality_mode (str): Determines how this component fits the seasonality. Options are "additive" and "multiplicative". Defaults to "additive".
stan_backend (str): Determines the backend that should be used to run Prophet. Options are "CMDSTANPY" and "PYSTAN". Defaults to "CMDSTANPY".
interval_width (float): Determines the confidence of the prediction interval range when calling `get_prediction_intervals`.
Accepts values in the range (0,1). Defaults to 0.95.
random_seed (int): Seed for the random number generator. Defaults to 0.
"""
name = "Prophet Regressor"
hyperparameter_ranges = {
"changepoint_prior_scale": Real(0.001, 0.5),
"seasonality_prior_scale": Real(0.01, 10),
"holidays_prior_scale": Real(0.01, 10),
"seasonality_mode": ["additive", "multiplicative"],
}
"""{
"changepoint_prior_scale": Real(0.001, 0.5),
"seasonality_prior_scale": Real(0.01, 10),
"holidays_prior_scale": Real(0.01, 10),
"seasonality_mode": ["additive", "multiplicative"],
}"""
model_family = ModelFamily.PROPHET
"""ModelFamily.PROPHET"""
supported_problem_types = [ProblemTypes.TIME_SERIES_REGRESSION]
"""[ProblemTypes.TIME_SERIES_REGRESSION]"""
def __init__(
self,
time_index: Optional[Hashable] = None,
changepoint_prior_scale: float = 0.05,
seasonality_prior_scale: int = 10,
holidays_prior_scale: int = 10,
seasonality_mode: str = "additive",
stan_backend: str = "CMDSTANPY",
interval_width: float = 0.95,
random_seed: Union[int, float] = 0,
**kwargs,
):
parameters = {
"changepoint_prior_scale": changepoint_prior_scale,
"seasonality_prior_scale": seasonality_prior_scale,
"holidays_prior_scale": holidays_prior_scale,
"seasonality_mode": seasonality_mode,
"stan_backend": stan_backend,
"interval_width": interval_width,
}
parameters.update(kwargs)
c_error_msg = (
"cmdstanpy is not installed. Please install using `pip install cmdstanpy`."
)
_ = import_or_raise("cmdstanpy", error_msg=c_error_msg)
p_error_msg = (
"prophet is not installed. Please install using `pip install prophet`."
)
prophet = import_or_raise("prophet", error_msg=p_error_msg)
prophet_regressor = prophet.Prophet(**parameters)
parameters["time_index"] = time_index
self.time_index = time_index
super().__init__(
parameters=parameters,
component_obj=prophet_regressor,
random_state=random_seed,
)
[docs] @staticmethod
def build_prophet_df(
X: pd.DataFrame,
y: Optional[pd.Series] = None,
time_index: str = "ds",
) -> pd.DataFrame:
"""Build the Prophet data to pass fit and predict on."""
X = copy.deepcopy(X)
y = copy.deepcopy(y)
if time_index is None:
raise ValueError("time_index cannot be None!")
if time_index in X.columns:
date_column = X.pop(time_index)
else:
raise ValueError(f"Column {time_index} was not found in X!")
prophet_df = X
if y is not None:
y.index = prophet_df.index
prophet_df["y"] = y
prophet_df["ds"] = date_column
return prophet_df
[docs] def fit(self, X: pd.DataFrame, y: Optional[pd.Series] = None):
"""Fits Prophet regressor component to data.
Args:
X (pd.DataFrame): The input training data of shape [n_samples, n_features].
y (pd.Series): The target training data of length [n_samples].
Returns:
self
"""
X, y = super()._manage_woodwork(X, y)
prophet_df = ProphetRegressor.build_prophet_df(
X=X,
y=y,
time_index=self.time_index,
)
self._component_obj.fit(prophet_df)
return self
[docs] def predict(self, X: pd.DataFrame, y: Optional[pd.Series] = None) -> pd.Series:
"""Make predictions using fitted Prophet regressor.
Args:
X (pd.DataFrame): Data of shape [n_samples, n_features].
y (pd.Series): Target data. Ignored.
Returns:
pd.Series: Predicted values.
"""
X = infer_feature_types(X)
prophet_df = ProphetRegressor.build_prophet_df(
X=X,
y=y,
time_index=self.time_index,
)
prophet_output = self._component_obj.predict(prophet_df)
predictions = prophet_output["yhat"]
predictions = infer_feature_types(predictions)
predictions = predictions.rename(None)
predictions.index = X.index
return predictions
[docs] def get_prediction_intervals(
self,
X: pd.DataFrame,
y: Optional[pd.Series] = None,
coverage: List[float] = None,
predictions: pd.Series = None,
) -> Dict[str, pd.Series]:
"""Find the prediction intervals using the fitted ProphetRegressor.
Args:
X (pd.DataFrame): Data of shape [n_samples, n_features].
y (pd.Series): Target data. Ignored.
coverage (List[float]): A list of floats between the values 0 and 1 that the upper and lower bounds of the
prediction interval should be calculated for.
predictions (pd.Series): Not used for Prophet estimator.
Returns:
dict: Prediction intervals, keys are in the format {coverage}_lower or {coverage}_upper.
"""
if coverage is None:
coverage = [0.95]
prophet_df = ProphetRegressor.build_prophet_df(
X=X,
y=y,
time_index=self.time_index,
)
prediction_interval_result = {}
for conf_int in coverage:
self._component_obj.interval_width = conf_int
X = infer_feature_types(X)
prophet_output = self._component_obj.predict(prophet_df)
prophet_output.index = X.index
prediction_interval_lower = prophet_output["yhat_lower"]
prediction_interval_upper = prophet_output["yhat_upper"]
prediction_interval_result[f"{conf_int}_lower"] = prediction_interval_lower
prediction_interval_result[f"{conf_int}_upper"] = prediction_interval_upper
return prediction_interval_result
[docs] def get_params(self) -> dict:
"""Get parameters for the Prophet regressor."""
return self.__dict__["_parameters"]
@property
def feature_importance(self) -> np.ndarray:
"""Returns array of 0's with len(1) as feature_importance is not defined for Prophet regressor."""
return np.zeros(1)
@classproperty
def default_parameters(cls) -> dict:
"""Returns the default parameters for this component.
Returns:
dict: Default parameters for this component.
"""
parameters = {
"changepoint_prior_scale": 0.05,
"time_index": None,
"seasonality_prior_scale": 10,
"holidays_prior_scale": 10,
"interval_width": 0.95,
"seasonality_mode": "additive",
"stan_backend": "CMDSTANPY",
}
return parameters
