Univariate Prophet

Bases: BaseProphetForecaster

Univariate Prophetverse forecaster, with support for multiple likelihoods.

Differences to facebook's prophet:

• logistic trend. Here, another parametrization is considered, and the capacity is not passed as input, but inferred from the data.

• the users can pass arbitrary sktime transformers as feature_transformer, for instance FourierFeatures or HolidayFeatures.

• no default weekly_seasonality/yearly_seasonality, this is left to the user via the feature_transformer parameter

• Uses changepoint_interval instead of n_changepoints to set changepoints.

• accepts configurations where each exogenous variable has a different function relating it to its additive effect on the time series. One can, for example, set different priors for a group of feature, or use a Hill function to model the effect of a feature.

Parameters:

Name	Type	Description	Default
trend	Union[str, BaseEffect], optional, one of "linear" (default) or "logistic"	Type of trend to use. Can also be a custom effect object.	'linear'
changepoint_interval	int	Number of potential changepoints to sample in the history.	25
changepoint_range	float or int	Proportion of the history in which trend changepoints will be estimated. if float, must be between 0 and 1. The range will be that proportion of the training history. if int, can be positive or negative. Absolute value must be less than number of training points. The range will be that number of points. A negative int indicates number of points counting from the end of the history, a positive int from the beginning.	0.8
changepoint_prior_scale	float	Regularization parameter controlling the flexibility of the automatic changepoint selection.	0.001
offset_prior_scale	float	Scale parameter for the prior distribution of the offset. The offset is the constant term in the piecewise trend equation.	0.1
capacity_prior_scale	float	Scale parameter for the prior distribution of the capacity.	0.2
capacity_prior_loc	float	Location parameter for the prior distribution of the capacity.	1.1
feature_transformer	sktime transformer, BaseTransformer	Transformer object to generate Fourier terms, holiday or other features. If None, no additional features are used. For multiple features, pass a FeatureUnion object with the transformers.	None
noise_scale	float	Scale parameter for the observation noise.	0.05
mcmc_samples	int	Number of MCMC samples to draw.	2000
mcmc_warmup	int	Number of MCMC warmup steps. Also known as burn-in.	200
mcmc_chains	int	Number of MCMC chains to run in parallel.	4
inference_method	str, optional, one of "mcmc" or "map"	Inference method to use. Can be "mcmc" or "map".	"map"
optimizer_name	str	Name of the numpyro optimizer to use for variational inference.	"Adam"
optimizer_kwargs	dict	Additional keyword arguments to pass to the numpyro optimizer.	{}
optimizer_steps	int	Number of optimization steps to perform for variational inference.	100_000
exogenous_effects	List[AbstractEffect]	A list of prophetverse AbstractEffect objects defining the exogenous effects to be used in the model.	None
likelihood	str	Likelihood to use for the model. Can be "normal", "gamma" or "negbinomial".	"normal"
default_effect	AbstractEffectm optional, defalut=None	The default effect to be used when no effect is specified for a variable.	None
default_exogenous_prior	tuple	Default prior distribution for exogenous effects.	None
rng_key	jax.random.PRNGKey or None (default	Random number generator key.	None

Source code in src/prophetverse/sktime/univariate.py

class Prophetverse(BaseProphetForecaster):
    """Univariate ``Prophetverse`` forecaster, with support for multiple likelihoods.

    Differences to facebook's prophet:

    * logistic trend. Here, another parametrization is considered,
      and the capacity is not passed as input, but inferred from the data.

    * the users can pass arbitrary ``sktime`` transformers as ``feature_transformer``,
      for instance ``FourierFeatures`` or ``HolidayFeatures``.

    * no default weekly_seasonality/yearly_seasonality, this is left to the user
      via the ``feature_transformer`` parameter

    * Uses ``changepoint_interval`` instead of ``n_changepoints`` to set changepoints.

    * accepts configurations where each exogenous variable has a different function
      relating it to its additive effect on the time series.
      One can, for example, set different priors for a group of feature,
      or use a Hill function to model the effect of a feature.

    Parameters
    ----------
    trend : Union[str, BaseEffect], optional, one of "linear" (default) or "logistic"
        Type of trend to use. Can also be a custom effect object.

    changepoint_interval : int, optional, default=25
        Number of potential changepoints to sample in the history.

    changepoint_range : float or int, optional, default=0.8
        Proportion of the history in which trend changepoints will be estimated.

        * if float, must be between 0 and 1.
          The range will be that proportion of the training history.

        * if int, can be positive or negative.
          Absolute value must be less than number of training points.
          The range will be that number of points.
          A negative int indicates number of points
          counting from the end of the history, a positive int from the beginning.

    changepoint_prior_scale : float, optional, default=0.001
        Regularization parameter controlling the flexibility
        of the automatic changepoint selection.

    offset_prior_scale : float, optional, default=0.1
        Scale parameter for the prior distribution of the offset.
        The offset is the constant term in the piecewise trend equation.

    capacity_prior_scale : float, optional, default=0.2
        Scale parameter for the prior distribution of the capacity.

    capacity_prior_loc : float, optional, default=1.1
        Location parameter for the prior distribution of the capacity.

    feature_transformer : sktime transformer, BaseTransformer, optional, default=None
        Transformer object to generate Fourier terms, holiday or other features.
        If None, no additional features are used.
        For multiple features, pass a ``FeatureUnion`` object with the transformers.

    noise_scale : float, optional, default=0.05
        Scale parameter for the observation noise.

    mcmc_samples : int, optional, default=2000
        Number of MCMC samples to draw.

    mcmc_warmup : int, optional, default=200
        Number of MCMC warmup steps. Also known as burn-in.

    mcmc_chains : int, optional, default=4
        Number of MCMC chains to run in parallel.

    inference_method : str, optional, one of "mcmc" or "map", default="map"
        Inference method to use. Can be "mcmc" or "map".

    optimizer_name : str, optional, default="Adam"
        Name of the numpyro optimizer to use for variational inference.

    optimizer_kwargs : dict, optional, default={}
        Additional keyword arguments to pass to the numpyro optimizer.

    optimizer_steps : int, optional, default=100_000
        Number of optimization steps to perform for variational inference.

    exogenous_effects : List[AbstractEffect], optional, default=None
        A list of ``prophetverse`` ``AbstractEffect`` objects
        defining the exogenous effects to be used in the model.

    likelihood : str, optional, default="normal"
        Likelihood to use for the model. Can be "normal", "gamma" or "negbinomial".

    default_effect : AbstractEffectm optional, defalut=None
        The default effect to be used when no effect is specified for a variable.

    default_exogenous_prior : tuple, default=None
        Default prior distribution for exogenous effects.

    rng_key : jax.random.PRNGKey or None (default
        Random number generator key.
    """

    _tags = {
        # packaging info
        # --------------
        "authors": "felipeangelimvieira",
        "maintainers": "felipeangelimvieira",
        "python_dependencies": "prophetverse",
        # estimator type
        # --------------
        "capability:pred_int": True,
        "capability:pred_int:insample": True,
        "enforce_index_type": [pd.Period, pd.DatetimeIndex],
        "requires-fh-in-fit": False,
        "y_inner_mtype": "pd.DataFrame",
    }

    def __init__(
        self,
        trend: Union[BaseEffect, str] = "linear",
        changepoint_interval: int = 25,
        changepoint_range: Union[float, int] = 0.8,
        changepoint_prior_scale: float = 0.001,
        offset_prior_scale: float = 0.1,
        capacity_prior_scale=0.2,
        capacity_prior_loc=1.1,
        exogenous_effects: Optional[List[BaseEffect]] = None,
        default_effect: Optional[BaseEffect] = None,
        feature_transformer=None,
        noise_scale=0.05,
        mcmc_samples=2000,
        mcmc_warmup=200,
        mcmc_chains=4,
        inference_method="map",
        optimizer_name="Adam",
        optimizer_kwargs=None,
        optimizer_steps=100_000,
        likelihood="normal",
        scale=None,
        rng_key=None,
    ):
        """Initialize the Prophet model."""
        self.noise_scale = noise_scale
        self.feature_transformer = feature_transformer

        self.likelihood = likelihood

        super().__init__(
            rng_key=rng_key,
            # Trend
            trend=trend,
            changepoint_interval=changepoint_interval,
            changepoint_range=changepoint_range,
            changepoint_prior_scale=changepoint_prior_scale,
            offset_prior_scale=offset_prior_scale,
            capacity_prior_scale=capacity_prior_scale,
            capacity_prior_loc=capacity_prior_loc,
            # Exog
            default_effect=default_effect,
            exogenous_effects=exogenous_effects,
            # BaseBayesianForecaster
            inference_method=inference_method,
            mcmc_samples=mcmc_samples,
            mcmc_warmup=mcmc_warmup,
            mcmc_chains=mcmc_chains,
            optimizer_name=optimizer_name,
            optimizer_kwargs=optimizer_kwargs,
            optimizer_steps=optimizer_steps,
            scale=scale,
        )

        self._validate_hyperparams()

    @property
    def model(self):
        """Return the model function.

        Returns
        -------
        Callable
            A function that can be passed to Numpyro samplers.
        """
        return _LIKELIHOOD_MODEL_MAP[self.likelihood]

    @property
    def _likelihood_is_discrete(self) -> bool:
        """Skip scaling if the likelihood is discrete.

        In the case of discrete likelihoods, the data is not scaled since this can
        create non-integer values.
        """
        return self.likelihood in _DISCRETE_LIKELIHOODS

    def _validate_hyperparams(self):
        """Validate the hyperparameters."""
        super()._validate_hyperparams()

        if self.noise_scale <= 0:
            raise ValueError("noise_scale must be greater than 0.")

        if self.likelihood not in _LIKELIHOOD_MODEL_MAP:
            raise ValueError(
                f"likelihood must be one of {list(_LIKELIHOOD_MODEL_MAP.keys())}"
                + "Got {self.likelihood}."
            )

    def _get_fit_data(self, y, X, fh):
        """
        Prepare the data for the Numpyro model.

        Parameters
        ----------
        y: pd.DataFrame
            Time series data.
        X: pd.DataFrame
            Exogenous variables.
        fh: ForecastingHorizon
            Forecasting horizon.

        Returns
        -------
        dict
            Dictionary of data for the Numpyro model.
        """
        fh = y.index.get_level_values(-1).unique()

        self.trend_model_ = self._get_trend_model()

        if self._likelihood_is_discrete:
            # Scale the data, since _get_fit_data receives
            # a non-scaled y for discrete likelihoods
            self.trend_model_.fit(X=X, y=y / self._scale)
        else:
            self.trend_model_.fit(X=X, y=y)

        # Exogenous features

        if X is None:
            X = pd.DataFrame(index=y.index)

        if self.feature_transformer is not None:

            X = self.feature_transformer.fit_transform(X)

        self._has_exogenous = ~X.columns.empty
        X = X.loc[y.index]

        trend_data = self.trend_model_.transform(X=X, fh=fh)

        self._fit_effects(X, y)
        exogenous_data = self._transform_effects(X, fh=fh)

        y_array = jnp.array(y.values.flatten()).reshape((-1, 1))

        # Inputs that also are used in predict
        self.fit_and_predict_data_ = {
            "trend_model": self.trend_model_,
            "noise_scale": self.noise_scale,
            "scale": self._scale,
            "exogenous_effects": (
                self.non_skipped_exogenous_effect if self._has_exogenous else None
            ),
        }

        inputs = {
            "y": y_array,
            "data": exogenous_data,
            "trend_data": trend_data,
            **self.fit_and_predict_data_,
        }

        return inputs

    def _get_predict_data(
        self, X: Union[pd.DataFrame, None], fh: ForecastingHorizon
    ) -> dict:
        """
        Prepare the data for making predictions.

        Parameters
        ----------
        X: pd.DataFrame
            Exogenous variables.
        fh: ForecastingHorizon
            Forecasting horizon.

        Returns
        -------
        dict
            Dictionary of data for the Numpyro model.
        """
        fh_dates = self.fh_to_index(fh)
        fh_as_index = pd.Index(list(fh_dates.to_numpy()))

        if X is None:
            X = pd.DataFrame(index=fh_as_index)

        if self.feature_transformer is not None:
            X = self.feature_transformer.transform(X)

        trend_data = self.trend_model_.transform(X=X, fh=fh_as_index)

        exogenous_data = (
            self._transform_effects(X, fh_as_index) if self._has_exogenous else None
        )

        return dict(
            y=None,
            data=exogenous_data,
            trend_data=trend_data,
            **self.fit_and_predict_data_,
        )

    @classmethod
    def get_test_params(cls, parameter_set="default"):  # pragma: no cover
        """Params to be used in sktime unit tests.

        Parameters
        ----------
        parameter_set : str, optional
            The parameter set to be used (ignored in this implementation)

        Returns
        -------
        List[dict[str, int]]
            A list of dictionaries containing the test parameters.
        """
        params = []
        for likelihood in _LIKELIHOOD_MODEL_MAP.keys():
            params.append(
                {
                    "likelihood": likelihood,
                    "optimizer_steps": 10,
                }
            )
        return params

model property

Return the model function.

Returns:

Type	Description
Callable	A function that can be passed to Numpyro samplers.

__init__(trend='linear', changepoint_interval=25, changepoint_range=0.8, changepoint_prior_scale=0.001, offset_prior_scale=0.1, capacity_prior_scale=0.2, capacity_prior_loc=1.1, exogenous_effects=None, default_effect=None, feature_transformer=None, noise_scale=0.05, mcmc_samples=2000, mcmc_warmup=200, mcmc_chains=4, inference_method='map', optimizer_name='Adam', optimizer_kwargs=None, optimizer_steps=100000, likelihood='normal', scale=None, rng_key=None)

Initialize the Prophet model.

Source code in src/prophetverse/sktime/univariate.py

def __init__(
    self,
    trend: Union[BaseEffect, str] = "linear",
    changepoint_interval: int = 25,
    changepoint_range: Union[float, int] = 0.8,
    changepoint_prior_scale: float = 0.001,
    offset_prior_scale: float = 0.1,
    capacity_prior_scale=0.2,
    capacity_prior_loc=1.1,
    exogenous_effects: Optional[List[BaseEffect]] = None,
    default_effect: Optional[BaseEffect] = None,
    feature_transformer=None,
    noise_scale=0.05,
    mcmc_samples=2000,
    mcmc_warmup=200,
    mcmc_chains=4,
    inference_method="map",
    optimizer_name="Adam",
    optimizer_kwargs=None,
    optimizer_steps=100_000,
    likelihood="normal",
    scale=None,
    rng_key=None,
):
    """Initialize the Prophet model."""
    self.noise_scale = noise_scale
    self.feature_transformer = feature_transformer

    self.likelihood = likelihood

    super().__init__(
        rng_key=rng_key,
        # Trend
        trend=trend,
        changepoint_interval=changepoint_interval,
        changepoint_range=changepoint_range,
        changepoint_prior_scale=changepoint_prior_scale,
        offset_prior_scale=offset_prior_scale,
        capacity_prior_scale=capacity_prior_scale,
        capacity_prior_loc=capacity_prior_loc,
        # Exog
        default_effect=default_effect,
        exogenous_effects=exogenous_effects,
        # BaseBayesianForecaster
        inference_method=inference_method,
        mcmc_samples=mcmc_samples,
        mcmc_warmup=mcmc_warmup,
        mcmc_chains=mcmc_chains,
        optimizer_name=optimizer_name,
        optimizer_kwargs=optimizer_kwargs,
        optimizer_steps=optimizer_steps,
        scale=scale,
    )

    self._validate_hyperparams()

get_test_params(parameter_set='default') classmethod

Params to be used in sktime unit tests.

Parameters:

Name	Type	Description	Default
parameter_set	str	The parameter set to be used (ignored in this implementation)	'default'

Returns:

Type	Description
List[dict[str, int]]	A list of dictionaries containing the test parameters.

Source code in src/prophetverse/sktime/univariate.py

@classmethod
def get_test_params(cls, parameter_set="default"):  # pragma: no cover
    """Params to be used in sktime unit tests.

    Parameters
    ----------
    parameter_set : str, optional
        The parameter set to be used (ignored in this implementation)

    Returns
    -------
    List[dict[str, int]]
        A list of dictionaries containing the test parameters.
    """
    params = []
    for likelihood in _LIKELIHOOD_MODEL_MAP.keys():
        params.append(
            {
                "likelihood": likelihood,
                "optimizer_steps": 10,
            }
        )
    return params