Calculates improvement in the loss function compared to the predictions of a null model. Can be used to evaluate model quality with respect to

Takes a scorer and returns a scorer that ignores NA / infinite elements in y_true. sklearn scorers (and others) don't handle arrays with 0 length

Aggregates input array. Divides array from left to right into bins of size agg_periods, and applies agg_func to each block. Drops records from

Checks if the ``freq`` parameter passed to a function has unit at most "D" Parameters ---------- freq : `DateOffset`, `Timedelta` or `str

Returns all interactions between static_col and fourier series up to specified order :param static_col: column to interact with fourier s

Given two lists of anomaly ranges, calculate a range-based score over the time series represented by the first list of anomaly ranges. If ``anomal

Fits predictive ML (machine learning) models to continuous response vector (given in ``y_col``) and returns fitted model. Parameters

Returns the fraction of predicted values whose relative difference from the true value is strictly greater than ``rtol``. Parameters ----

Generic test for the results of any given scenario

Returns aggregated data. Parameters ---------- df : `pandas.DataFrame`. The input data must have TIME_COL ("ts")

Automatically group holidays and their neighboring days based on estimated holiday impact. Parameters ---------- df : `pandas.DataFrame`

Returns the number of training points in `df`, the start year, and prediction end year Parameters ---------- df : `pandas.DataFrame` with

Returns the number of splitting iterations in the cross-validator as configured, ignoring self.max_splits and self.min_splits Par

Returns the largest timestamp that is smaller than or equal to `ts` and is also a multiple of the `freq`. Assume hourly frequency i.e. `freq`

Infers frequency of the timestamps provided in the ``time_col`` of ``df``. Notes ----- If the timeseries does not have any missing values

Evaluates the logistic function at x with the specified growth rate, capacity, floor, and inflection point. :param x: value to evaluate t

For a given dataframe with columns including numerical values, it generates a function which can be applied to original data as well as any fu

Computes the precision scores for two arrays. Parameters ---------- y_true : array-like, 1-D The actual categories. y_pred :

Makes prediction for a given x. Parameters ---------- X : `numpy.array`, `pandas.DataFrame` or `pandas.Series` Th

Predicts the quantiles of the ``offset_col`` (defined in ``ci_model``) in ``new_df``. Parameters ---------- new_df : `pd.Dataframe`

Converts an instance of the `ADConfig` class to its dictionary format.

Returns the unique elements in the input list, preserving the original order. Parameters ---------- array: `List` [`any`]

Fetches pipeline parameters from the ``df`` and ``config``, but does not run the pipeline to generate a forecast. :py:meth:`~greykite

Asserts that the two pipelines are equal The Pipelines should be created by `get_basic_pipeline`.

Asserts whether CV splits match expectation :param actual: np.array of np.arrays :param expected: np.array of np.arrays

Uses information in `ADConfig` to construct a reward function. The constructed reward function will be the sum of various rewards related to t

Helper function to create evaluation plots from traces. Creates a figure with subplots. Every dataframe in ``traces`` has the same columns.

Filters change points that are too close. Given the ``cp_block`` from the output of ``find_neighbor_changepoints`` and the corresponding coef

Returns holidays within the countries between ``year_start`` and ``year_end``. Creates a separate key, value for each item in ``holidays_to_model_

This function identifies extreme values as outliers based on z-scores. A normal distribution will be fit on ``value_col`` of input df, and the

Gets the coefficient summary filtered by conditions. Parameters ---------- is_intercept : `bool` or `None`, default `None`

Returns default forecast horizon based on input data period and num_observations :param period: float Period of each observation (i.e. ave

Returns hyperparameter grid. Implements the method in `~greykite.framework.templates.base_template.BaseTemplate`. Converts model com

Positional reward for a single pointwise anomaly with an anomaly range Parameters ---------- loc : `int` Location of the pointwis

Loads dataset by name from the internal data library. Parameters ---------- data_name : `str` Dataset to load fro

Returns categorical term for patsy. Optionally specify levels, coding, and quote the term :param term: str name of the categorical

``predict`` method is already implemented here and should work for most cases. This is because in most cases, ``predict`` is simply

Gets the column name for a specific stage/frequency. The name will be f"{self.time_col_}__{index}__{freq}". Parameters ------

Takes a scorer and returns a scorer that pre-aggregates input. :param score_func: function that maps two arrays to a number. E.g. (y_true, y_pred

Checks whether evaluation functions return the same output for a few inputs

A function which returns a dataframe including aggregated (e.g. averaged) time series lags in the form of dataframe columns. By "aggregated la

A function to generate yearly seasonality features from a given time series df. The seasonality features include n_changepoints * 2 columns of th

Returns a serialized string representation of the `obj`. The `obj` must be serializable by `dill`. Serialized output `dill` is a byte

Fits ``ARIMA`` forecast model. Parameters ---------- X : `pandas.DataFrame` Input timeseries, with timestamp colu

The fit method for Greykite based method. The fit consists of following stages: - Fit a forecast model using the ``forecast_confi

Fits the partial regularized regression. Parameters ---------- x : `numpy.array` or `pandas.DataFrame` The design

Generates a func which adds multiple fourier series with multiple periods. Parameters ---------- col_names : `list` [`str`] the c

Converts a dictionary to the corresponding instance of the `ADConfig` class. Raises ValueError if the input is not a dictionary.

Generates holidays for Prophet model. Parameters ---------- year_list : `list` [`int`] List of years for selectin

Returns a diagonal weight matrix with shape (``n_forecasts``, ``n_forecasts``) and Frobenius norm sqrt(`n_forecasts`). Parameters -------

This is just a helper function to generate plots for outlier detection during tests. This plots the raw input and marks the discovered anomalies.

Makes a plot to show the observations/estimations/change points. In this function, component parameters specify if each component in the plot

Utility function which overlayes the predicted anomalies or anomalies on the forecast vs actual plot. The function calls the internal function `~g

A function which plots ``actual_col`` and ``forecast_col`` and well as ``forecast_lower_col``, ``forecast_upper_col`` with respect to x axis

Convert a list of pointwise anomalies into a list of anomaly ranges Parameters ---------- pointwise_anmalies : array-like, 1-D Li

Orders columns according to ``order_dict``. Can be used to order columns according to hierarchical constraints. Consider the tree where a par

Partitions `n` elements into adjacent groups, each with `group_size` elements Group number starts from 0 and increments upward

Returns unique elements in ``array``, removing all levels of nesting if found. Parameters ---------- array : `list` [`any`] or None

Validates the inputs to the class for the method ``run``. Raises a ValueError if there is a mismatch between the following parameters

Checks that necessary input are provided in ``self.uncertainty_dict`` and ``self.train_df``. This check only raises `~greykite.sklearn

Sets the derived attributes from model init parameters.

Checks if a time series consists of equal time increments and if it is in increasing order. :param df: data.frame which includes the time col

Evaluates the adjustment quality. Computes the following metrics for each of the `m` timeseries: "Base MAPE" : MAPE of ba

Generates silverkite forecast breakdown for groupings given in ``grouping_regex_patterns_dict``. Note that this only works for additiv

A wrapper around ``SilverkiteForecast::forecast`` that simplifies some of the input parameters. Parameters ---------- args :

Gets the most appropriate model template that fits the input df's frequency, forecast horizon and number of cv splits. We define the cv to be

The functions takes the ``changepoints_dict`` dictionary and returns the processed ``changepoints_dict``. If "method" == "auto", the change p

Partitions interval into n_changepoints + 1 segments, placing a changepoint at left endpoint of each segment. The left most segment do

Gets the distance from end of time series where no changepoints will be placed. Parameters ---------- min_increment : `datetime.timedelta

Loads the Beijing Particulate Matter (PM2.5) dataset. https://archive.ics.uci.edu/ml/datasets/Beijing+PM2.5+Data This hourly data set

Calculating min and max lag order needed given the prescribed lags in the model. :param lag_dict: dict dictionary with these fields:

A function which applies to a potentially new data frame (``new_df``) with the same columns as ``df`` (different values or row number is allow

Makes a plot of the observed data and estimated components, as well as detected changes The function currently allows five different components t

Plots forecast with prediction intervals, against actuals Adapted from plotly user guide: https://plot.ly/python/v3/continuous-error-bars/#bas

Creates forecast for dates specified in X Parameters ---------- X : pd.DataFrame with column (self.time_col_), the dates for

Performs predictions and calculate uncertainty using multiple simulations. Parameters ---------- fut_df : `pandas.Dat

A function to simulate future series. If the fitted model supports uncertainty e.g. via ``uncertainty_dict``, errors are incorporated

Appends the resample time columns to ``df``. For example, the original df has hourly data with columns "ts" and "y". The original tim

Samples up to ``max_splits`` items from list(range(`num_splits`)). If ``use_most_recent_splits`` is True, highest split indices up to ``max_s

For each key of event_df_dict, it adds a new column to a data frame (df) with a date column (date_col). Each new column will represent the eve

Calculates a baseline for the input series `y` and then removes that baseline from `y`, thus creating a diff series. Parameters

Expands an input holiday dataframe ``holiday_df`` to include the neighboring days specified in ``holiday_impact_dict`` or through ``holiday_im

Computes the F1 scores for two arrays. Parameters ---------- y_true : array-like, 1-D The actual categories. y_pred : array-l

Fits a model to training data. Parameters ---------- X : `pandas.DataFrame` Input timeseries, with timestamp colu

Fits the regressor using X[value_col] Parameters ---------- X : pd.DataFrame with columns time_col and value_col

Automatically generates trend changepoint detection parameters based on the input data and forecast horizon. Parameters ---------- df

Returns the values of continuous_time_col at the requested changepoint_dates. :param df: pd.DataFrame training dataset. contains

Returns the names of the available internal datasets. Returns ------- file_names : `list` [`str`] The names of th

Calculates overlap reward for both size and position of two anomaly ranges Parameters ---------- anomaly_range_1 : list [int] A l

For a given frequency, it returns a lag which is likely to be most correlated to the observation at current time. For daily data, this will r

Validate ``config_names`` against keys of ``configs``. Raises a ValueError in case of a mismatch. Parameters ----------

Estimates the impact of holidays and their neighboring days and groups events with similar effects to several groups using kernel density esti

A function to impute timeseries values (given in ``df``) and in ``value_col`` with chosen lagged values or an aggregated of those. For example

Loads the Daily Peyton Manning dataset. This dataset contains log daily page views for the Wikipedia page for Peyton Manning. One of

Returns a deserialized object from a `serialized_string`, usually the output of `dumps_to_str`. Parameters ----------

Calculates the mean interval score. If an observed value falls within the interval, the score is simply the width of the interval. If an obser

Evaluates the objective function for a given transform matrix and dataset. Parameters ---------- transform_matrix

Predicts the interval. Parameters ---------- fut_df : `pandas.DataFrame` The dataframe used for prediction.