Adds integer index value (position) to the provided `index`. For attribute=='relative', the reference point/index is the prediction/forecast i

Single Encoders must implement an _encode() method to encode the index. Parameters ---------- index The index gen

Extracts and returns the target components from an input batch Parameters ---------- batch The input batch tuple for the forward

Convert different `axis` types to an integer axis.

Computes the error scale based on a naive seasonal forecasts on `insample` values with seasonality `m`.

Returns the encoder or decoder variable of the TFT model. Parameters ---------- weights The weights of the encode

Given two overlapping series `ts_target` and `ts_covariate` and an index point `idx` of `ts_target`, returns the matching index point in `ts_

Computes normalized tolerance levels.

Create a representation of the TimeSeries values. The returned dimensions respect the maximum allowed items to be displayed Paramete

Darts model is only used to detect which covariates are supported by the weights.

Launches the encode sequence for past covariates and future covariates for either training, inference/prediction or training and inference/pre

Loads a checkpoint for the underlying :class:`PLForecastingModule` (PLM) model. The PLM object is not stored when saving a :class:`TorchForeca

Plots the variable importance of the TFT model. Parameters ---------- importance The encoder / decoder importance

Returns the TimeSeries linearly reconciled from the projection matrix G and the summation matrix S.

Expands a tensor `x` of shape (batch size, n components) to a tensor of shape (batch size, `ocl`, n target components, 1 (n samples)), by repeatin

Return `times` resampled with frequency `freq` and values with `np.nan` for the newly inserted dates. The frequency `freq` must represent a t

Divides ``errors`` by ``scale``, handling zero-scale entries gracefully. When ``zero_division`` is ``"warn"`` (default), the behavior depends on

Select the absolute number of samples randomly if an integer has been supplied. If a float has been supplied, select a fraction

Find the start (absolute index) and end (index relative to the end) indices that represent the time intersection from `self` and `other`.

Load the ILINetDataset dataset as a list of univariate timeseries.

Validates the model.

r""" Creates the features array `X` and labels array `y` to train a lagged-variables `SKLearnModel` when `is_training = True`; alternatively,

Returns a new TimeSeries with index `time_index` and one or more dimensions containing (optionally one-hot encoded or cyclic encoded) pd.Data

Describe one or more options. Available Options: - display.[max_rows, max_cols] - plotting.use_darts_style Parameters ----

Computes a score/metric between predicted anomalies against true anomalies. `pred_anomalies` and `anomalies` must have: - identical dime

Partitions the series into a sequence of sub-series by using significant gaps of missing values Parameters ---------- series

First values of the potentially multivariate series. Returns ------- numpy.ndarray The first values of every comp

Fit the model to data, separately for each output variable. Parameters ---------- X : {array-like, sparse matrix} of shape (n

Fits the likelihood to the model. Parameters ---------- model The model to fit the likelihood to. The mo

Fit/train the model on the provided series. Parameters ---------- series A target time series. The model will be

Create a ``TimeSeries`` from a CSV file. One column can be used to represent the time (if not present, the time index will be a RangeIndex)

Generates covariate encodings for training and inference/prediction and returns a tuple of past, and future covariates series with the origina

Generates covariate encodings for the inference/prediction set and returns a tuple of past, and future covariates series with the original and

Returns the time-dependent attention on the encoder and decoder for each `horizon` in (1, `output_chunk_length`). The time index rang

Helper function that returns all the times within `past_covariates` that can be used to create features for training or prediction.

checks that covariates generated by encoders fulfill the requirements compared to some reference covariates: What has to match:

Generates expected prediction results for naive conformal model from: - residuals and predictions from deterministic/probabilistic model

Infers the frequency at which two frequencies `freq` and `other` intersect. Parameters ---------- freq The first frequency. o

Loads a model from a given file path. Example for loading a general save from :class:`RNNModel`: .. highlight:: python

Load the model configuration from a JSON file. Returns ------- dict The model configuration.

Load the model by creating an instance of the given module class and loading the weights. Some configuration files might contain external para

Return the largest slice of the deterministic series without any gaps (contiguous all-NaN value entries) larger than `max_gap_size`.

Plots data relevant to residuals. This function takes a univariate TimeSeries instance of residuals and plots their values, their distributio

Aggregates the (sequence of) multivariate binary series given as input into a (sequence of) univariate binary series. Parameters

Compute the anomaly score(s) for the given (sequence of) series. Predicts the given target time series with the filtering model, and applies

Set the value of an option.

Plot the results of the scorer. Computes the anomaly score on the two series. And plots the results. The plot will be composed of th

Return the time index of the series where the time index was intersected with the `other` series. This method is in general *not* symmetric.

Parameters ---------- max_slope The slope of the steeper parallelogram side.

MAE loss as defined in: https://en.wikipedia.org/wiki/Mean_absolute_error. Given a time series of actual values :math:`y_t` and a ti

Parameters ---------- window Integer value indicating the size of the window W used by the scorer to transform th

Recurrent Neural Network Model (RNNs). This class provides three variants of RNNs: * Vanilla RNN * LSTM * GRU

Temporal Convolutional Network Model (TCN). This is an implementation of a dilated TCN used for forecasting, inspired from [1]_. Thi

Block Recurrent Neural Network Model (RNNs). This is a neural network model that uses an RNN encoder to encode fixed-length input chunks, and

An implementation of the TiDE model, as presented in [1]_. TiDE is similar to Transformers (implemented in :class:`TransformerModel`),

Transformer model Transformer is a state-of-the-art deep learning model introduced in 2017. It is an encoder-decoder architecture who

PatchTST-FM Model for zero-shot forecasting. This is an implementation of IBM's PatchTST-FM model [1]_, ported from `ibm-granite/gran

Base Conformal Prediction Model. Base class for any conformal prediction model. A conformal model calibrates the predictions from any

Shrink the beginning and end of the historical forecasts time index based on the value of `start`.

If relative start position `rel_start` is negative, then adjust it to the first non-negative index that lies a round-multiple of `stride` ahead of

Checks that `fit_called` is `True`.

Models can override this method to return a custom `TorchTrainingDataset`.

Generate calibrated historical forecasts. In general the workflow of the models to produce one calibrated forecast/prediction per step in the

Checks if the scorer has been fitted before calling its `score()` function.