Convert kwargs to dictionary, check that their content is compatible with called methods.

Initializes a PyTorch-Lightning trainer for training or prediction from `trainer_params`.

Return the encoders from a model save with several sanity checks.

Given a Path to the file and a DataLoaderMetadata object, return a TimeSeries One can assume that the file exists and its MD5 checksu

Makes attention mask.

Build an additive attention mask of shape (B, Q, K) from query/key padding masks. Parameters ---------- query_pad (B, Q) bool or

Build base models for combination tests. Returns a list of forecasting models, optionally pre-fitted when ``train_fcm=False``.

For GlobalForecastingModel, when predicting n > output_chunk_length, `historical_forecasts()` generally produce better forecasts than

Slides a window of size `window` along the input series, and replaces the value of the input time series by the maximum of the values contained in

Prepend the historic part of the `insample` series to the `series` if it is not None.

Wrapper around model inference method

This is the basic way to compute the NLL loss. It can be overwritten by likelihoods for which PyTorch proposes a numerically better N

Return all observed/inferred frequencies of a `pandas.DatetimeIndex`. The frequencies are inferred from all combinations of three consecutive

Optimized historical forecasts for TorchForecastingModels Rely on _check_optimizable_historical_forecasts() to check that the assumptions ar

Optimized historical forecasts for SKLearnModel. Rely on _check_optimizable_historical_forecasts() to check that the assumptions are verifie

Helper function to plot series and anomalies. Parameters ---------- series The actual series to visualize anomalies from. ano

Helper function, for consistency with the other trends

Generate predictions. Generates deterministic predictions if no `Likelihood` was used. Otherwise, generates probabilistic predictions

Forecasts values for a certain number of time steps after the end of the series. DualCovariatesModels must implement the predict logic in this

Aggregates the sequence of multivariate binary series given as input into a sequence of univariate binary series. assuming the input is

Processes input and returns two lists of tuples `(encoder_id, attribute)` from relevant encoder parameters at model creation. Paramet

Processes input params used at model creation and returns tuple of one transformer object and two masks that specify which past / future encod

Extracts numerical and categorical static covariate (component / columns) names and their component masks in order of the input data.

If static covariates are component-specific, they must be reshaped appropriately.

Processes input params used at model creation for time zone specification, and returns the time zone. Parameters ---------- p

Generates train output. Feeds `PLForecastingModule` with (past target + past cov + historic future cov (concatenated), future cov, st

Helper that extracts and returns the explainability result attribute for a given component. Parameters ---------- at

Handles the regression metrics input parameters and checks.

Return `times` re-indexed into a `pandas.RangeIndex` and `values` in the re-indexed order. An integer `pandas.Index` can be converted to a `p

For each timestamp, the corresponding distribution is fitted on the probabilistic time-series input_2, and returns the negative log-likelihood

SDPA attention implementation using torch.nn.functional.scaled_dot_product_attention. Args: query_states: [batch, n_heads, seq_le

Sets up/Initializes all past and future encoders and an optional transformer from `add_encoder` parameter used at model creation. Pa

Sets up the model for fine-tuning based on `self.enable_finetuning`.

Sets up/Initializes an optional transformer from `add_encoder` parameter used at model creation. Parameters ---------- params

Returns stochastic forecast of `n_samples` samples. This method is a replicate of Prophet.predict() which suspends simplification of stochasti

Computes the tolerance coverages for different tolerance levels. More info in metric `autc()`.

Performs the actual training Parameters ---------- train_loader the training data loader feeding the tra

Returns unique labels for each component in the true and predicted labels.

Unpack the dictionary that is returned by the StatsForecast 'predict()' method. Into an array of quantile predictions with shape (n (horizon), n

if user supplied additional covariates to model.fit() or model.predict(), `self.transform_mask` has to be updated as user-defined covariates s

Updates the running stats.

Check if categorical features are integer-encoded

Helper that validates the input parameters of a method that queries the `ComponentBasedExplainabilityResult`. Parameters ---

Helper that validates the input parameters of a method that queries the `HorizonBasedExplainabilityResult`. Parameters -----

validate that parameters used at model creation are part of the model cls __init__, its parents __init__ methods, or :class:`PLForecastingModu

Verify the `enable_finetuning` input.

Verify that the provided inference dataset is of the correct type

Verify that the provided train dataset is of the correct type

Applies Rotary Position Embedding to the query and key tensors. Args: q (`torch.Tensor`): The query tensor. k (`torch

Gives the number of active grid cells in a column. Parameters ---------- column A column in the window, must be w

Decode patches through transformer and quantile head.

Describe option(s) matching the pattern.

The dtype of the series' values.

Returns a tuple of (past covariates encoders, future covariates encoders)

Defines how to ensemble the individual models' predictions to produce a single prediction. Parameters ---------- pre

Fits the likelihood to the model.

Trains the detector on the given time series. Parameters ---------- series Time (sequence of) series to be used t

Fit the aggregators on the (sequence of) multivariate binary anomaly series. If a list of series is given, they must have the same number of

Fit/train the model on one or multiple series. This method wraps around :func:`fit_from_dataset()`, constructing a default training d