TinyTimeMixerForecaster#

class TinyTimeMixerForecaster(model_path='ibm/TTM', revision='main', validation_split=0.2, config=None, training_args=None, compute_metrics=None, callbacks=None, broadcasting=False, use_source_package=False, fit_strategy='minimal', device='cpu', freq=None, verbose=False, padding_mask='observed')[source]#

TinyTimeMixer Forecaster for Zero-Shot Forecasting of Multivariate Time Series.

Wrapping implementation in [1] of method proposed in [2]. See [3] for tutorial by creators.

TinyTimeMixer (TTM) are compact pre-trained models for Time-Series Forecasting, open-sourced by IBM Research. With less than 1 Million parameters, TTM introduces the notion of the first-ever “tiny” pre-trained models for Time-Series Forecasting.

Fit Strategies: Full, Minimal, and Zero-shot

This model supports three fit strategies: zero-shot for direct predictions without training, minimal fine-tuning for lightweight adaptation to new data, and full fine-tuning for comprehensive model training. The selected strategy is determined by the model’s fit_strategy parameter

Initialization Process:

  1. Model Path: The model_path parameter points to a local folder or Hugging Face repo that contains both configuration files and pretrained weights. Public TinyTimeMixer checkpoints are available for the R1 [4], R2 [5], research R2 [6], and R3 [7] model families.

  2. Default Configuration: The model loads its default configuration from the configuration files.

  3. Custom Configuration: Users can provide a custom configuration via the config parameter during model initialization.

  4. Configuration Override: If custom configuration is provided, it overrides the default configuration.

  5. Forecasting Horizon: If revision=None, the forecasting horizon (fh) specified during fit is used to select a compatible checkpoint revision. For prediction, the requested horizon must fit within the loaded model’s config.prediction_length.

  6. Model Architecture: The final configuration is used to construct the model architecture.

  7. Pretrained Weights: pretrained weights are loaded from the model_path, these weights are then aligned and loaded into the model architecture.

  8. Weight Alignment: However sometimes, pretrained weights do not align with the model architecture, because the config was changed which created a model architecture of different size than the default one. This causes some of the weights in model architecture to be reinitialized randomly instead of using the pre-trained weights.

Training Strategies:

  • Zero-shot Forecasting: When all the pre-trained weights are correctly aligned with the model architecture, fine-tuing part is bypassed and the model preforms zero-short forecasting.

  • Minimal Fine-tuning: When not all the pre-trained weights are correctly aligned with the model architecture, rather some weights are re-initialized, these re-initialized weights are fine-tuned on the provided data.

  • Full Fine-tuning: The model is fully fine-tuned on new data, updating all parameters. This approach offers maximum adaptation to the dataset but requires more computational resources.

Exogenous Variables Support

TTM supports exogenous variables (external factors) that can improve forecasting accuracy. The model accepts exogenous variables that are known for both the historical period and the future forecasting horizon.

When using exogenous variables: - The X parameter should contain exogenous data covering both

past and future periods

  • Exogenous variables must have the same index structure as the target series

  • For prediction, exogenous data must extend into the forecasting horizon

Parameters:
model_pathstr, default=”ibm/TTM”

Path to the Hugging Face model to use for forecasting. This can be either:

  • The name of a Hugging Face repository, for example "ibm-research/ttm-r3" [7]. Related checkpoints are listed in the TinyTimeMixer [8], Granite time series [9], and IBM Research time series [10] collections.

  • A local path to a folder containing model files in a format supported by transformers. In this case, ensure that the directory contains all necessary files (e.g., configuration, tokenizer, and model weights).

  • If this parameter is None, fit_strategy should be full to allow training a randomly initialized model from the provided or default config, else ValueError is raised.

revisionstr or None, default=”main”

Revision of the model to use:

  • None: Automatically select a compatible revision based on the training context length and forecasting horizon.

  • “main”: Load the main branch of the selected checkpoint.

  • A checkpoint branch name such as “52-16-ft-r2.1” can be used to load a specific TinyTimeMixer variant.

This param becomes irrelevant when model_path is None.

devicestr, default=”cpu”

Device for model inference and fine-tuning, for example "cpu", "cuda", "cuda:0", or "auto". "auto" is passed to transformers device_map and selects an available accelerator.

freqstr or None, default=None

Frequency to pass to models that use resolution prefix tuning, such as TTM-R2 [5] and research R2 [6]. If None, the frequency is inferred from the forecasting horizon or time index where possible, and falls back to the out-of-vocabulary token.

verbosebool, default=False

If True, show training output from transformers.Trainer.

validation_splitfloat, default=0.2

Fraction of the data to use for validation

configdict or None, default={}

Configuration to use for the model. See the transformers documentation for details. The provided configuration must be valid for the selected TinyTimeMixer model architecture. Configuration inherits from transformers.PretrainedConfig and can be used to control the model outputs.

context_length (int, optional, defaults to 64)

The context/history length for the input sequence.

patch_length (int, optional, defaults to 8)

The patch length for the input sequence.

num_input_channels (int):

Number of input variates. For Univariate, set it to 1.

patch_stride (int, optional, defaults to 8):

Amount of points to stride. If its value is same as patch_length, we get non-overlapping patches.

d_model (int, optional, defaults to 16):

Hidden feature size of the model.

prediction_length (int, optional, defaults to 16)

Number of time steps to forecast for a forecasting task. Also known as the Forecast Horizon.

num_parallel_samples (int, optional, defaults to 100):

The number of samples to generate in parallel for probabilistic forecast.

expansion_factor (int, optional, defaults to 2):

Expansion factor to use inside MLP. Recommended range is 2-5. Larger value indicates more complex model.

num_layers (int, optional, defaults to 3):

Number of layers to use. Recommended range is 3-15. Larger value indicates more complex model.

dropout (float, optional, defaults to 0.2):

The dropout probability the TinyTimeMixer backbone. Recommended range is 0.2-0.7

mode (str, optional, defaults to “common_channel”):

Mixer Mode. Determines how to process the channels. Allowed values: “common_channel”, “mix_channel”. In “common_channel” mode, we follow Channel-independent modelling with no explicit channel-mixing. Channel mixing happens in an implicit manner via shared weights across channels. (preferred first approach) In “mix_channel” mode, we follow explicit channel-mixing in addition to patch and feature mixer. (preferred approach when channel correlations are very important to model)

gated_attn (bool, optional, defaults to True):

Enable Gated Attention.

norm_mlp (str, optional, defaults to “LayerNorm”):

Normalization layer (BatchNorm or LayerNorm).

self_attn (bool, optional, defaults to False):

Enable Tiny self attention across patches. This can be enabled when the output of Vanilla TinyTimeMixer with gated attention is not satisfactory. Enabling this leads to explicit pair-wise attention and modelling across patches.

self_attn_heads (int, optional, defaults to 1):

Number of self-attention heads. Works only when self_attn is set to True.

use_positional_encoding (bool, optional, defaults to False):

Enable the use of positional embedding for the tiny self-attention layers. Works only when self_attn is set to True.

positional_encoding_type (str, optional, defaults to “sincos”):

Positional encodings. Options “random” and “sincos” are supported. Works only when use_positional_encoding is set to True

scaling (string or bool, optional, defaults to “std”):

Whether to scale the input targets via “mean” scaler, “std” scaler or no scaler if None. If True, the scaler is set to “mean”.

loss (string, optional, defaults to “mse”):

The loss function to finetune or pretrain the the model. Allowed values are “mse” or “mae” or “pinball” or “huber”. Use pinball loss for probabilistic forecasts of different quantiles. Distribution head (nll) is currently disabled and not allowed.

init_std (float, optional, defaults to 0.02):

The standard deviation of the truncated normal weight initialization distribution.

post_init (bool, optional, defaults to False):

Whether to use custom weight initialization from transformers library, or the default initialization in PyTorch. Setting it to False performs PyTorch weight initialization.

norm_eps (float, optional, defaults to 1e-05):

A value added to the denominator for numerical stability of normalization.

adaptive_patching_levels (int, optional, defaults to 0):

If adaptive_patching_levels is i, then we will have i levels with each level having n_layers. Level id starts with 0. num_patches at level i will be multipled by (2^i) and num_features at level i will be divided by (2^i). For Ex. if adaptive_patching_levels is 3 - then we will have 3 levels:

  • level 2: num_features//(2^2), num_patches*(2^2)

  • level 1: num_features//(2^1), num_patches*(2^1)

  • level 0: num_features//(2^0), num_patches*(2^0)

adaptive_patching_levels = 1 is same as one level PatchTSMixer. This module gets disabled when adaptive_patching_levels is 0 or neg value. Defaults to 0 (off mode).

resolution_prefix_tuning (bool, optional, defaults to False):

Enable if your dataloader has time resolution information as defined in get_freq_mapping function in modelling_tinytimemixer.

frequency_token_vocab_size (int, optional, defaults to 5):

Vocab size to use when resolution_prefix_tuning is enabled.

head_dropout (float, optional, defaults to 0.2):

The dropout probability the TinyTimeMixer head.

distribution_output (string, optional, defaults to “student_t”):

The distribution emission head for the model when loss is “nll”. Could be either “student_t”, “normal” or “negative_binomial”.

prediction_channel_indices (list, optional):

List of channel indices to forecast. If None, forecast all channels. Target data is expected to have all channels and we explicitly filter the channels in prediction and target before loss computation. Please provide the indices in sorted ascending order.

exogenous_channel_indices (list, optional):

List of channel indices whose values are known in the forecast period. Please provide the indices in sorted ascending order.

decoder_num_layers (int, optional, defaults to 8):

Number of layers to use in decoder

decoder_d_model(int, optional, defaults to 16):

Defines the hidden feature size of the decoder.

decoder_adaptive_patching_levels (int, optional, defaults to 0):

Adaptive Patching levels for decoder. Preferable to set it to 0 for decoder to keep it light weight.

decoder_raw_residual (bool, optional, defaults to False):

Flag to enable merging of raw embedding with encoder embedding for decoder input. Defaults to False.

decoder_mode (string, optional, defaults to “common_channel”):

Decoder channel mode. Use “common_channel” for channel-independent modelling and `”mix_channel” for channel-mixing modelling

use_decoder (bool, optional, defaults to True):

Enable to use decoder.

enable_forecast_channel_mixing (bool, optional, defaults to False):

Enable if we want to reconcile forecasts across all channels and also to enable exogenous infusion, if you have them.

fcm_gated_attn (bool, optional, defaults to True):

Enable gated attention in forecast channel mixing block.

fcm_context_length (int, optional, defaults to `1):

Surrounding context length to use. For Ex. If we want to consider 2 lag point before and after a data point, provide value 2 for fcm_context_length

fcm_use_mixer (bool, optional, defaults to True):

Enable Mixing in forecast channel mixing block.

fcm_mix_layers (int, optional, defaults to 2):

Number of mixer layers to use if fcm_use_mixer is enabled

fcm_prepend_past (bool, optional, defaults to True):

Prepend last context for forecast reconciliation

fcm_prepend_past_offset (int, optional, defaults to None): categorical_vocab_size_list (list, optional):

List of vocab size for all the tokenized categorical variables to use. Pass it in the same order as used in the foreward call param static_categorical_values.

prediction_filter_length (int,*optional*, defaults to None):

Actual length in the prediction output to use for loss calculations.

training_argsdict or None, default={}

Training arguments to use for the model. See transformers.TrainingArguments for details [11]. Note that the output_dir argument is required.

compute_metricslist, default=[]

List of metrics to compute during training. See transformers.Trainer for details.

callbackslist, default=None

List of callbacks to use during training. See transformers.Trainer

broadcastingbool, default=False

if True, multiindex data input will be broadcasted to single series. For each single series, one copy of this forecaster will try to fit and predict on it. The broadcasting is happening inside automatically, from the outerside api perspective, the input and output are the same, only one multiindex output from predict.

use_source_packagebool, default=False

If True, the model and configuration will be loaded directly from the source package tsfm_public.models.tinytimemixer. This is useful if you want to bypass the local version of the package or when working in an environment where the latest updates from the source package are needed. If False, the model and configuration will be loaded from the local version of package maintained in sktime because of model’s unavailability on pypi. To install the source package, follow the instructions here [1].

fit_strategystr, default=”minimal”

Strategy to use for fitting (fine-tuning) the model. This can be one of the following: - “zero-shot”: Uses pre-trained model as it is. If model path is None

with this strategy, ValueError is raised.

  • “minimal”: Fine-tunes only a small subset of the model parameters, allowing for quick adaptation with limited computational resources. If model path is None with this strategy, ValueError is raised.

  • “full”: Fine-tunes all model parameters, which may result in better performance but requires more computational power and time. Allows model path to be None.

padding_maskstr, default=”observed”

Controls how synthetic padding is masked when the input history is shorter than the model’s context_length. Missing history is left-padded with zeros. This can be one of the following:

  • “observed”: Marks the padded zeros as observed, matching Granite-TSFM’s preprocessing (past_observed_mask built with ~np.isnan, so numeric zeros count as observed). Pretrained TinyTimeMixer checkpoints were trained under these semantics, so this is required to reproduce Granite-TSFM’s forecasts and is the only valid option when fit_strategy="zero-shot".

  • “unobserved”: Marks the padded zeros as unobserved, so they are not treated as part of the observed series. This does not match how public checkpoints were pretrained, and is only valid when fit_strategy is "minimal" or "full", so the model can be fine-tuned under these mask semantics; otherwise a ValueError is raised.

Attributes:
cutoff

Cut-off = “present time” state of forecaster.

fh

Forecasting horizon that was passed.

is_fitted

Whether fit has been called.

state

State of the estimator.

References

Examples

Zero-shot forecasting with a pretrained TinyTimeMixer R3 checkpoint [7]. Other supported public checkpoints include R1 [4], R2 [5], and research R2 [6] variants:

>>> from sktime.forecasting.ttm import TinyTimeMixerForecaster
>>> from sktime.datasets import load_airline
>>> y = load_airline()
>>> forecaster = TinyTimeMixerForecaster(
...     model_path="ibm-research/ttm-r3",
...     # Other supported public checkpoints include:
...     # model_path="ibm-granite/granite-timeseries-ttm-r1",
...     # model_path="ibm-granite/granite-timeseries-ttm-r2",
...     # model_path="ibm-granite/granite-timeseries-ttm-r2",
...     # revision="52-16-ft-r2.1",
...     # model_path="ibm-research/ttm-research-r2",
... )
>>> forecaster.fit(y)
TinyTimeMixerForecaster(...)
>>> y_pred = forecaster.predict(fh=[1, 2, 3])

Automatically select the best compatible model revision from the context length and prediction length:

>>> from sktime.datasets import load_airline
>>> from sktime.forecasting.ttm import TinyTimeMixerForecaster
>>> y = load_airline()
>>> forecaster = TinyTimeMixerForecaster(
...     model_path="ibm-research/ttm-r3",
...     revision=None,
... )
>>> forecaster.fit(y, fh=[1, 2, 3])
TinyTimeMixerForecaster(...)
>>> y_pred = forecaster.predict()

Forecasting with exogenous variables known during training and prediction:

>>> from sktime.forecasting.ttm import TinyTimeMixerForecaster
>>> from sktime.datasets import load_longley
>>> from sktime.split import temporal_train_test_split
>>> y, X = load_longley()
>>> y_train, _, X_train, X_future = temporal_train_test_split(y, X, test_size=2)
>>> forecaster = TinyTimeMixerForecaster(
...     model_path="ibm-research/ttm-r3",
... )
>>> forecaster.fit(y_train, X=X_train, fh=[1, 2])
TinyTimeMixerForecaster(...)
>>> y_pred = forecaster.predict(X=X_future)

Minimal fine-tuning updates only parameters that are not loaded from the checkpoint, for example when the supplied configuration changes the model shape:

>>> from sktime.datasets import load_airline
>>> from sktime.forecasting.ttm import TinyTimeMixerForecaster
>>> y = load_airline()
>>> forecaster = TinyTimeMixerForecaster(
...     model_path="ibm-research/ttm-r3",
...     fit_strategy="minimal",
...     config={
...         "context_length": 24,
...         "trend_patch_length": 6,
...         "trend_patch_stride": 6,
...         "prediction_length": 12,
...     },
...     training_args={
...         "max_steps": 10,
...         "output_dir": "test_output",
...         "per_device_train_batch_size": 4,
...         "report_to": "none",
...     },
... )
>>> forecaster.fit(y)
TinyTimeMixerForecaster(...)
>>> y_pred = forecaster.predict(fh=[1, 2, 3])

Initialize a random model when model_path is None and preform full fine-tuning:

>>> from sktime.datasets import load_airline
>>> from sktime.forecasting.ttm import TinyTimeMixerForecaster
>>> y = load_airline()
>>> forecaster = TinyTimeMixerForecaster(
...     model_path=None,
...     fit_strategy="full",
...     training_args={
...         "max_steps": 10,
...         "output_dir": "test_output",
...         "per_device_train_batch_size": 4,
...         "report_to": "none",
...     },
... )
>>> forecaster.fit(y)
TinyTimeMixerForecaster(...)
>>> y_pred = forecaster.predict(fh=[1, 2, 3])

Pretrain on panel data before fitting to the target forecasting series:

>>> from sktime.forecasting.ttm import TinyTimeMixerForecaster
>>> from sktime.datasets import load_airline, load_tecator
>>> y = load_airline()
>>> y_panel = load_tecator(
...     return_type="pd-multiindex",
...     return_X_y=False,
... )
>>> y_panel.drop(["class_val"], axis=1, inplace=True)
>>> forecaster = TinyTimeMixerForecaster(
...     model_path="ibm-research/ttm-r3",
...     fit_strategy="full",
...     training_args={
...         "max_steps": 10,
...         "output_dir": "test_output",
...         "per_device_train_batch_size": 4,
...         "report_to": "none",
...     },
... )
>>> forecaster.pretrain(y_panel)
TinyTimeMixerForecaster(...)
>>> forecaster.fit(y)
TinyTimeMixerForecaster(...)
>>> y_pred = forecaster.predict(fh=[1, 2, 3])

Methods

check_is_fitted([method_name])

Check if the estimator has been fitted.

clone()

Obtain a clone of the object with same hyper-parameters and config.

clone_tags(estimator[, tag_names])

Clone tags from another object as dynamic override.

create_test_instance([parameter_set])

Construct an instance of the class, using first test parameter set.

create_test_instances_and_names([parameter_set])

Create list of all test instances and a list of names for them.

fit(y[, X, fh])

Fit forecaster to training data.

fit_predict(y[, X, fh, X_pred])

Fit and forecast time series at future horizon.

get_class_tag(tag_name[, tag_value_default])

Get class tag value from class, with tag level inheritance from parents.

get_class_tags()

Get class tags from class, with tag level inheritance from parent classes.

get_config()

Get config flags for self.

get_fitted_params([deep])

Get fitted parameters.

get_param_defaults()

Get object's parameter defaults.

get_param_names([sort])

Get object's parameter names.

get_params([deep])

Get a dict of parameters values for this object.

get_pretrained_params([deep])

Get pretrained parameters of this estimator.

get_tag(tag_name[, tag_value_default, ...])

Get tag value from instance, with tag level inheritance and overrides.

get_tags()

Get tags from instance, with tag level inheritance and overrides.

get_test_params([parameter_set])

Return testing parameter settings for the estimator.

is_composite()

Check if the object is composed of other BaseObjects.

load_from_path(serial)

Load object from file location.

load_from_serial(serial)

Load object from serialized memory container.

predict([fh, X])

Forecast time series at future horizon.

predict_interval([fh, X, coverage])

Compute/return prediction interval forecasts.

predict_proba([fh, X, marginal])

Compute/return fully probabilistic forecasts.

predict_quantiles([fh, X, alpha])

Compute/return quantile forecasts.

predict_residuals([y, X])

Return residuals of time series forecasts.

predict_var([fh, X, cov])

Compute/return variance forecasts.

pretrain(y[, X, fh])

Pre-train forecaster on panel (global) data.

reset()

Reset the object to a clean post-init state.

save([path, serialization_format])

Save serialized self to bytes-like object or to (.zip) file.

score(y[, X, fh])

Scores forecast against ground truth, using MAPE (non-symmetric).

set_config(**config_dict)

Set config flags to given values.

set_params(**params)

Set the parameters of this object.

set_random_state([random_state, deep, ...])

Set random_state pseudo-random seed parameters for self.

set_tags(**tag_dict)

Set instance level tag overrides to given values.

update(y[, X, update_params])

Update cutoff value and, optionally, fitted parameters.

update_predict(y[, cv, X, update_params, ...])

Make predictions and update model iteratively over the test set.

update_predict_single([y, fh, X, update_params])

Update model with new data and make forecasts.

classmethod get_test_params(parameter_set='default')[source]#

Return testing parameter settings for the estimator.

Parameters:
parameter_setstr, default=”default”

Name of the set of test parameters to return, for use in tests. If no special parameters are defined for a value, will return “default” set. There are currently no reserved values for forecasters.

Returns:
paramsdict or list of dict, default = {}

Parameters to create testing instances of the class Each dict are parameters to construct an “interesting” test instance, i.e., MyClass(**params) or MyClass(**params[i]) creates a valid test instance. create_test_instance uses the first (or only) dictionary in params

check_is_fitted(method_name=None)[source]#

Check if the estimator has been fitted.

Check if _is_fitted attribute is present and True. The is_fitted attribute should be set to True in calls to an object’s fit method.

If not, raises a NotFittedError.

Parameters:
method_namestr, optional

Name of the method that called this function. If provided, the error message will include this information.

Raises:
NotFittedError

If the estimator has not been fitted yet.

clone()[source]#

Obtain a clone of the object with same hyper-parameters and config.

A clone is a different object without shared references, in post-init state. This function is equivalent to returning sklearn.clone of self.

Equivalent to constructing a new instance of type(self), with parameters of self, that is, type(self)(**self.get_params(deep=False)).

If configs were set on self, the clone will also have the same configs as the original, equivalent to calling cloned_self.set_config(**self.get_config()).

Also equivalent in value to a call of self.reset, with the exception that clone returns a new object, instead of mutating self like reset.

Raises:
RuntimeError if the clone is non-conforming, due to faulty __init__.
clone_tags(estimator, tag_names=None)[source]#

Clone tags from another object as dynamic override.

Every scikit-base compatible object has a dictionary of tags. Tags may be used to store metadata about the object, or to control behaviour of the object.

Tags are key-value pairs specific to an instance self, they are static flags that are not changed after construction of the object.

clone_tags sets dynamic tag overrides from another object, estimator.

The clone_tags method should be called only in the __init__ method of an object, during construction, or directly after construction via __init__.

The dynamic tags are set to the values of the tags in estimator, with the names specified in tag_names.

The default of tag_names writes all tags from estimator to self.

Current tag values can be inspected by get_tags or get_tag.

Parameters:
estimatorAn instance of :class:BaseObject or derived class
tag_namesstr or list of str, default = None

Names of tags to clone. The default (None) clones all tags from estimator.

Returns:
self

Reference to self.

classmethod create_test_instance(parameter_set='default')[source]#

Construct an instance of the class, using first test parameter set.

Parameters:
parameter_setstr, default=”default”

Name of the set of test parameters to return, for use in tests. If no special parameters are defined for a value, will return “default” set.

Returns:
instanceinstance of the class with default parameters
classmethod create_test_instances_and_names(parameter_set='default')[source]#

Create list of all test instances and a list of names for them.

Parameters:
parameter_setstr, default=”default”

Name of the set of test parameters to return, for use in tests. If no special parameters are defined for a value, will return “default” set.

Returns:
objslist of instances of cls

i-th instance is cls(**cls.get_test_params()[i])

nameslist of str, same length as objs

i-th element is name of i-th instance of obj in tests. The naming convention is {cls.__name__}-{i} if more than one instance, otherwise {cls.__name__}

property cutoff[source]#

Cut-off = “present time” state of forecaster.

Returns:
cutoffpandas compatible index element, or None

pandas compatible index element, if cutoff has been set; None otherwise

property fh[source]#

Forecasting horizon that was passed.

fit(y, X=None, fh=None)[source]#

Fit forecaster to training data.

State change:

Changes state to “fitted”.

Writes to self:

  • Sets fitted model attributes ending in “_”, fitted attributes are inspectable via get_fitted_params.

  • Sets self.is_fitted flag to True.

  • Sets self.cutoff to last index seen in y.

  • Stores fh to self.fh if fh is passed.

Parameters:
ytime series in sktime compatible data container format.

Time series to which to fit the forecaster.

Individual data formats in sktime are so-called mtype specifications, each mtype implements an abstract scitype.

  • Series scitype = individual time series, vanilla forecasting. pd.DataFrame, pd.Series, or np.ndarray (1D or 2D)

  • Panel scitype = collection of time series, global/panel forecasting. pd.DataFrame with 2-level row MultiIndex (instance, time), 3D np.ndarray (instance, variable, time), list of Series typed pd.DataFrame

  • Hierarchical scitype = hierarchical collection, for hierarchical forecasting. pd.DataFrame with 3 or more level row MultiIndex (hierarchy_1, ..., hierarchy_n, time)

For further details on data format, see glossary on mtype. For usage, see forecasting tutorial examples/01_forecasting.ipynb

fhint, list, pd.Index coercible, or ForecastingHorizon, default=None

The forecasting horizon encoding the time stamps to forecast at. If self.get_tag("requires-fh-in-fit") is True, must be passed in fit, not optional

Xtime series in sktime compatible format, optional (default=None).

Exogeneous time series to fit the model to. Should be of same scitype (Series, Panel, or Hierarchical) as y. If self.get_tag("X-y-must-have-same-index"), X.index must contain y.index.

Returns:
selfReference to self.
fit_predict(y, X=None, fh=None, X_pred=None)[source]#

Fit and forecast time series at future horizon.

Same as fit(y, X, fh).predict(X_pred). If X_pred is not passed, same as fit(y, fh, X).predict(X).

State change:

Changes state to “fitted”.

Writes to self:

  • Sets fitted model attributes ending in “_”, fitted attributes are inspectable via get_fitted_params.

  • Sets self.is_fitted flag to True.

  • Sets self.cutoff to last index seen in y.

  • Stores fh to self.fh.

Parameters:
ytime series in sktime compatible data container format

Time series to which to fit the forecaster.

Individual data formats in sktime are so-called mtype specifications, each mtype implements an abstract scitype.

  • Series scitype = individual time series, vanilla forecasting. pd.DataFrame, pd.Series, or np.ndarray (1D or 2D)

  • Panel scitype = collection of time series, global/panel forecasting. pd.DataFrame with 2-level row MultiIndex (instance, time), 3D np.ndarray (instance, variable, time), list of Series typed pd.DataFrame

  • Hierarchical scitype = hierarchical collection, for hierarchical forecasting. pd.DataFrame with 3 or more level row MultiIndex (hierarchy_1, ..., hierarchy_n, time)

For further details on data format, see glossary on mtype. For usage, see forecasting tutorial examples/01_forecasting.ipynb

fhint, list, pd.Index coercible, or ForecastingHorizon (not optional)

The forecasting horizon encoding the time stamps to forecast at.

If fh is not None and not of type ForecastingHorizon it is coerced to ForecastingHorizon via a call to _check_fh. In particular, if fh is of type pd.Index it is coerced via ForecastingHorizon(fh, is_relative=False)

Xtime series in sktime compatible format, optional (default=None).

Exogeneous time series to fit the model to. Should be of same scitype (Series, Panel, or Hierarchical) as y. If self.get_tag("X-y-must-have-same-index"), X.index must contain y.index.

X_predtime series in sktime compatible format, optional (default=None)

Exogeneous time series to use in prediction. If passed, will be used in predict instead of X. Should be of same scitype (Series, Panel, or Hierarchical) as y in fit. If self.get_tag("X-y-must-have-same-index"), X.index must contain fh index reference.

Returns:
y_predtime series in sktime compatible data container format

Point forecasts at fh, with same index as fh. y_pred has same type as the y that has been passed most recently: Series, Panel, Hierarchical scitype, same format (see above)

classmethod get_class_tag(tag_name, tag_value_default=None)[source]#

Get class tag value from class, with tag level inheritance from parents.

Every scikit-base compatible object has a dictionary of tags. Tags may be used to store metadata about the object, or to control behaviour of the object.

Tags are key-value pairs specific to an instance self, they are static flags that are not changed after construction of the object.

The get_class_tag method is a class method, and retrieves the value of a tag taking into account only class-level tag values and overrides.

It returns the value of the tag with name tag_name from the object, taking into account tag overrides, in the following order of descending priority:

  1. Tags set in the _tags attribute of the class.

  2. Tags set in the _tags attribute of parent classes,

in order of inheritance.

Does not take into account dynamic tag overrides on instances, set via set_tags or clone_tags, that are defined on instances.

To retrieve tag values with potential instance overrides, use the get_tag method instead.

Parameters:
tag_namestr

Name of tag value.

tag_value_defaultany type

Default/fallback value if tag is not found.

Returns:
tag_value

Value of the tag_name tag in self. If not found, returns tag_value_default.

classmethod get_class_tags()[source]#

Get class tags from class, with tag level inheritance from parent classes.

Every scikit-base compatible object has a dictionary of tags. Tags may be used to store metadata about the object, or to control behaviour of the object.

Tags are key-value pairs specific to an instance self, they are static flags that are not changed after construction of the object.

The get_class_tags method is a class method, and retrieves the value of a tag taking into account only class-level tag values and overrides.

It returns a dictionary with keys being keys of any attribute of _tags set in the class or any of its parent classes.

Values are the corresponding tag values, with overrides in the following order of descending priority:

  1. Tags set in the _tags attribute of the class.

  2. Tags set in the _tags attribute of parent classes,

in order of inheritance.

Instances can override these tags depending on hyper-parameters.

To retrieve tags with potential instance overrides, use the get_tags method instead.

Does not take into account dynamic tag overrides on instances, set via set_tags or clone_tags, that are defined on instances.

For including overrides from dynamic tags, use get_tags.

Returns:
collected_tagsdict

Dictionary of tag name : tag value pairs. Collected from _tags class attribute via nested inheritance. NOT overridden by dynamic tags set by set_tags or clone_tags.

get_config()[source]#

Get config flags for self.

Configs are key-value pairs of self, typically used as transient flags for controlling behaviour.

get_config returns dynamic configs, which override the default configs.

Default configs are set in the class attribute _config of the class or its parent classes, and are overridden by dynamic configs set via set_config.

Configs are retained under clone or reset calls.

Returns:
config_dictdict

Dictionary of config name : config value pairs. Collected from _config class attribute via nested inheritance and then any overrides and new tags from _onfig_dynamic object attribute.

get_fitted_params(deep=True)[source]#

Get fitted parameters.

State required:

Requires state to be “fitted”.

Parameters:
deepbool, default=True

Whether to return fitted parameters of components.

  • If True, will return a dict of parameter name : value for this object, including fitted parameters of fittable components (= BaseEstimator-valued parameters).

  • If False, will return a dict of parameter name : value for this object, but not include fitted parameters of components.

Returns:
fitted_paramsdict with str-valued keys

Dictionary of fitted parameters, paramname : paramvalue keys-value pairs include:

  • always: all fitted parameters of this object, as via get_param_names values are fitted parameter value for that key, of this object

  • if deep=True, also contains keys/value pairs of component parameters parameters of components are indexed as [componentname]__[paramname] all parameters of componentname appear as paramname with its value

  • if deep=True, also contains arbitrary levels of component recursion, e.g., [componentname]__[componentcomponentname]__[paramname], etc

Notes

Pretrained parameters (set via pretrain) are excluded from the result. Use get_pretrained_params to retrieve those.

classmethod get_param_defaults()[source]#

Get object’s parameter defaults.

Returns:
default_dict: dict[str, Any]

Keys are all parameters of cls that have a default defined in __init__. Values are the defaults, as defined in __init__.

classmethod get_param_names(sort=True)[source]#

Get object’s parameter names.

Parameters:
sortbool, default=True

Whether to return the parameter names sorted in alphabetical order (True), or in the order they appear in the class __init__ (False).

Returns:
param_names: list[str]

List of parameter names of cls. If sort=False, in same order as they appear in the class __init__. If sort=True, alphabetically ordered.

get_params(deep=True)[source]#

Get a dict of parameters values for this object.

Parameters:
deepbool, default=True

Whether to return parameters of components.

  • If True, will return a dict of parameter name : value for this object, including parameters of components (= BaseObject-valued parameters).

  • If False, will return a dict of parameter name : value for this object, but not include parameters of components.

Returns:
paramsdict with str-valued keys

Dictionary of parameters, paramname : paramvalue keys-value pairs include:

  • always: all parameters of this object, as via get_param_names values are parameter value for that key, of this object values are always identical to values passed at construction

  • if deep=True, also contains keys/value pairs of component parameters parameters of components are indexed as [componentname]__[paramname] all parameters of componentname appear as paramname with its value

  • if deep=True, also contains arbitrary levels of component recursion, e.g., [componentname]__[componentcomponentname]__[paramname], etc

get_pretrained_params(deep=True)[source]#

Get pretrained parameters of this estimator.

Returns a dictionary of attributes that were set during pretrain. These are attributes ending in "_" that appeared on the estimator after pretrain was called, and are tracked separately from the fitted parameters set by fit.

State required:

Requires state to be "pretrained" or "fitted" (after pretraining). Returns an empty dict if the estimator has not been pretrained, or does not have the capability:pretrain tag.

Parameters:
deepbool, default=True

Whether to return pretrained parameters of nested estimators.

  • If True, will return a dict of parameter name : value for this object, including pretrained parameters of nested estimators.

  • If False, will return a dict of parameter name : value for this object, but not include pretrained parameters of nested estimators.

Returns:
paramsdict

Dictionary of pretrained parameter names mapped to their values. Keys are attribute names ending in "_" that were set during pretraining. Returns empty dict if estimator has not been pretrained.

If deep=True, also contains keys/value pairs of nested estimators’ pretrained parameters, indexed as [attrname]__[paramname].

See also

pretrain

Pretrain estimator on a collection of time series.

get_fitted_params

Get parameters set during fit.

Examples

>>> from sktime.forecasting.dummy_global import DummyGlobalForecaster
>>> from sktime.utils._testing.hierarchical import _make_hierarchical
>>> y_panel = _make_hierarchical(
...     hierarchy_levels=(3,), min_timepoints=12, max_timepoints=12
... )
>>> forecaster = DummyGlobalForecaster()
>>> forecaster.pretrain(y_panel)
DummyGlobalForecaster()
>>> params = forecaster.get_pretrained_params()
>>> sorted(params.keys())
['global_mean_', ..., 'n_pretrain_timepoints_']
get_tag(tag_name, tag_value_default=None, raise_error=True)[source]#

Get tag value from instance, with tag level inheritance and overrides.

Every scikit-base compatible object has a dictionary of tags. Tags may be used to store metadata about the object, or to control behaviour of the object.

Tags are key-value pairs specific to an instance self, they are static flags that are not changed after construction of the object.

The get_tag method retrieves the value of a single tag with name tag_name from the instance, taking into account tag overrides, in the following order of descending priority:

  1. Tags set via set_tags or clone_tags on the instance,

at construction of the instance.

  1. Tags set in the _tags attribute of the class.

  2. Tags set in the _tags attribute of parent classes,

in order of inheritance.

Parameters:
tag_namestr

Name of tag to be retrieved

tag_value_defaultany type, optional; default=None

Default/fallback value if tag is not found

raise_errorbool

whether a ValueError is raised when the tag is not found

Returns:
tag_valueAny

Value of the tag_name tag in self. If not found, raises an error if raise_error is True, otherwise it returns tag_value_default.

Raises:
ValueError, if raise_error is True.

The ValueError is then raised if tag_name is not in self.get_tags().keys().

get_tags()[source]#

Get tags from instance, with tag level inheritance and overrides.

Every scikit-base compatible object has a dictionary of tags. Tags may be used to store metadata about the object, or to control behaviour of the object.

Tags are key-value pairs specific to an instance self, they are static flags that are not changed after construction of the object.

The get_tags method returns a dictionary of tags, with keys being keys of any attribute of _tags set in the class or any of its parent classes, or tags set via set_tags or clone_tags.

Values are the corresponding tag values, with overrides in the following order of descending priority:

  1. Tags set via set_tags or clone_tags on the instance,

at construction of the instance.

  1. Tags set in the _tags attribute of the class.

  2. Tags set in the _tags attribute of parent classes,

in order of inheritance.

Returns:
collected_tagsdict

Dictionary of tag name : tag value pairs. Collected from _tags class attribute via nested inheritance and then any overrides and new tags from _tags_dynamic object attribute.

is_composite()[source]#

Check if the object is composed of other BaseObjects.

A composite object is an object which contains objects, as parameters. Called on an instance, since this may differ by instance.

Returns:
composite: bool

Whether an object has any parameters whose values are BaseObject descendant instances.

property is_fitted[source]#

Whether fit has been called.

Checks if the estimator state is “fitted”. Returns True only after a successful call to fit(), not after pretrain().

Returns:
bool

True if the estimator has been fitted, False otherwise.

classmethod load_from_path(serial)[source]#

Load object from file location.

Parameters:
serialresult of ZipFile(path).open(“object)
Returns:
deserialized self resulting in output at path, of cls.save(path)
classmethod load_from_serial(serial)[source]#

Load object from serialized memory container.

Parameters:
serial1st element of output of cls.save(None)
Returns:
deserialized self resulting in output serial, of cls.save(None)
predict(fh=None, X=None)[source]#

Forecast time series at future horizon.

State required:

Requires state to be “fitted”, i.e., self.is_fitted=True.

Accesses in self:

  • Fitted model attributes ending in “_”.

  • self.cutoff, self.is_fitted

Writes to self:

Stores fh to self.fh if fh is passed and has not been passed previously.

Parameters:
fhint, list, pd.Index coercible, or ForecastingHorizon, default=None

The forecasting horizon encoding the time stamps to forecast at. Should not be passed if has already been passed in fit. If has not been passed in fit, must be passed, not optional

If fh is not None and not of type ForecastingHorizon it is coerced to ForecastingHorizon via a call to _check_fh. In particular, if fh is of type pd.Index it is coerced via ForecastingHorizon(fh, is_relative=False)

Xtime series in sktime compatible format, optional (default=None)

Exogeneous time series to use in prediction. Should be of same scitype (Series, Panel, or Hierarchical) as y in fit. If self.get_tag("X-y-must-have-same-index"), X.index must contain fh index reference.

Returns:
y_predtime series in sktime compatible data container format

Point forecasts at fh, with same index as fh. y_pred has same type as the y that has been passed most recently: Series, Panel, Hierarchical scitype, same format (see above)

predict_interval(fh=None, X=None, coverage=0.9)[source]#

Compute/return prediction interval forecasts.

If coverage is iterable, multiple intervals will be calculated.

State required:

Requires state to be “fitted”, i.e., self.is_fitted=True.

Accesses in self:

  • Fitted model attributes ending in “_”.

  • self.cutoff, self.is_fitted

Writes to self:

Stores fh to self.fh if fh is passed and has not been passed previously.

Parameters:
fhint, list, pd.Index coercible, or ForecastingHorizon, default=None

The forecasting horizon encoding the time stamps to forecast at. Should not be passed if has already been passed in fit. If has not been passed in fit, must be passed, not optional

If fh is not None and not of type ForecastingHorizon, it is coerced to ForecastingHorizon internally (via _check_fh).

  • if fh is int or array-like of int, it is interpreted as relative horizon, and coerced to a relative ForecastingHorizon(fh, is_relative=True).

  • if fh is of type pd.Index, it is interpreted as an absolute horizon, and coerced to an absolute ForecastingHorizon(fh, is_relative=False).

Xtime series in sktime compatible format, optional (default=None)

Exogeneous time series to use in prediction. Should be of same scitype (Series, Panel, or Hierarchical) as y in fit. If self.get_tag("X-y-must-have-same-index"), X.index must contain fh index reference.

coveragefloat or list of float of unique values, optional (default=0.90)

nominal coverage(s) of predictive interval(s)

Returns:
pred_intpd.DataFrame

Prediction interval forecasts, with columns and rows as follows:

  • Column has multi-index: first level is variable name from y in fit, second level coverage fractions for which intervals were computed. in the same order as in input coverage. Third level is string “lower” or “upper”, for lower/upper interval end.

  • Row index is fh, with additional (upper) levels equal to instance levels, from y seen in fit, if y seen in fit was Panel or Hierarchical.

  • Entries are forecasts of lower/upper interval end, for var in col index, at nominal coverage in second col index, lower/upper depending on third col index, for the row index. Upper/lower interval end forecasts are equivalent to quantile forecasts at alpha = 0.5 - c/2, 0.5 + c/2 for c in coverage.

predict_proba(fh=None, X=None, marginal=True)[source]#

Compute/return fully probabilistic forecasts.

Note:

  • currently only implemented for Series (non-panel, non-hierarchical) y.

  • requires skpro installed for the distribution objects returned.

State required:

Requires state to be “fitted”, i.e., self.is_fitted=True.

Accesses in self:

  • Fitted model attributes ending in “_”.

  • self.cutoff, self.is_fitted

Writes to self:

Stores fh to self.fh if fh is passed and has not been passed previously.

Parameters:
fhint, list, pd.Index coercible, or ForecastingHorizon, default=None

The forecasting horizon encoding the time stamps to forecast at. Should not be passed if has already been passed in fit. If has not been passed in fit, must be passed, not optional

If fh is not None and not of type ForecastingHorizon, it is coerced to ForecastingHorizon internally (via _check_fh).

  • if fh is int or array-like of int, it is interpreted as relative horizon, and coerced to a relative ForecastingHorizon(fh, is_relative=True).

  • if fh is of type pd.Index, it is interpreted as an absolute horizon, and coerced to an absolute ForecastingHorizon(fh, is_relative=False).

Xtime series in sktime compatible format, optional (default=None)

Exogeneous time series to use in prediction. Should be of same scitype (Series, Panel, or Hierarchical) as y in fit. If self.get_tag("X-y-must-have-same-index"), X.index must contain fh index reference.

marginalbool, optional (default=True)

whether returned distribution is marginal by time index

Returns:
pred_distskpro BaseDistribution

predictive distribution if marginal=True, will be marginal distribution by time point if marginal=False and implemented by method, will be joint

predict_quantiles(fh=None, X=None, alpha=None)[source]#

Compute/return quantile forecasts.

If alpha is iterable, multiple quantiles will be calculated.

State required:

Requires state to be “fitted”, i.e., self.is_fitted=True.

Accesses in self:

  • Fitted model attributes ending in “_”.

  • self.cutoff, self.is_fitted

Writes to self:

Stores fh to self.fh if fh is passed and has not been passed previously.

Parameters:
fhint, list, pd.Index coercible, or ForecastingHorizon, default=None

The forecasting horizon encoding the time stamps to forecast at. Should not be passed if has already been passed in fit. If has not been passed in fit, must be passed, not optional

If fh is not None and not of type ForecastingHorizon, it is coerced to ForecastingHorizon internally (via _check_fh).

  • if fh is int or array-like of int, it is interpreted as relative horizon, and coerced to a relative ForecastingHorizon(fh, is_relative=True).

  • if fh is of type pd.Index, it is interpreted as an absolute horizon, and coerced to an absolute ForecastingHorizon(fh, is_relative=False).

Xtime series in sktime compatible format, optional (default=None)

Exogeneous time series to use in prediction. Should be of same scitype (Series, Panel, or Hierarchical) as y in fit. If self.get_tag("X-y-must-have-same-index"), X.index must contain fh index reference.

alphafloat or list of float of unique values, optional (default=[0.05, 0.95])

A probability or list of, at which quantile forecasts are computed.

Returns:
quantilespd.DataFrame

Quantile forecasts, with columns and rows as follows:

  • Column has multi-index: first level is variable name from y in fit, second level being the values of alpha passed to the function.

  • Row index is fh, with additional (upper) levels equal to instance levels, from y seen in fit, if y seen in fit was Panel or Hierarchical.

  • Entries are quantile forecasts, for var in col index, at quantile probability in second col index, for the row index.

predict_residuals(y=None, X=None)[source]#

Return residuals of time series forecasts.

Residuals will be computed for forecasts at y.index.

If fh must be passed in fit, must agree with y.index. If y is an np.ndarray, and no fh has been passed in fit, the residuals will be computed at a fh of range(len(y.shape[0]))

State required:

Requires state to be “fitted”. If fh has been set, must correspond to index of y (pandas or integer)

Accesses in self:

Fitted model attributes ending in “_”. self.cutoff, self._is_fitted

Writes to self:

Nothing.

Parameters:
ytime series in sktime compatible data container format

Time series with ground truth observations, to compute residuals to. Must have same type, dimension, and indices as expected return of predict.

Required; ground truth observations to compute residuals against.

Xtime series in sktime compatible format, optional (default=None)

Exogeneous time series for updating and forecasting Should be of same scitype (Series, Panel, or Hierarchical) as y in fit. If self.get_tag("X-y-must-have-same-index"), X.index must contain both fh index reference and y.index.

Returns:
y_restime series in sktime compatible data container format

Forecast residuals at fh`, with same index as ``fh. y_res has same type as the y that has been passed most recently: Series, Panel, Hierarchical scitype, same format (see above)

predict_var(fh=None, X=None, cov=False)[source]#

Compute/return variance forecasts.

State required:

Requires state to be “fitted”, i.e., self.is_fitted=True.

Accesses in self:

  • Fitted model attributes ending in “_”.

  • self.cutoff, self.is_fitted

Writes to self:

Stores fh to self.fh if fh is passed and has not been passed previously.

Parameters:
fhint, list, pd.Index coercible, or ForecastingHorizon, default=None

The forecasting horizon encoding the time stamps to forecast at. Should not be passed if has already been passed in fit. If has not been passed in fit, must be passed, not optional

If fh is not None and not of type ForecastingHorizon, it is coerced to ForecastingHorizon internally (via _check_fh).

  • if fh is int or array-like of int, it is interpreted as relative horizon, and coerced to a relative ForecastingHorizon(fh, is_relative=True).

  • if fh is of type pd.Index, it is interpreted as an absolute horizon, and coerced to an absolute ForecastingHorizon(fh, is_relative=False).

Xtime series in sktime compatible format, optional (default=None)

Exogeneous time series to use in prediction. Should be of same scitype (Series, Panel, or Hierarchical) as y in fit. If self.get_tag("X-y-must-have-same-index"), X.index must contain fh index reference.

covbool, optional (default=False)

currently unused, present for downwards compatibility and future extension

Returns:
pred_varpd.DataFrame, format dependent on cov variable

Variance forecasts, with columns and rows as follows:

  • Column names are exactly those of y passed in fit/update. For nameless formats, column index will be a RangeIndex.

  • Row index is fh, with additional levels equal to instance levels, from y seen in fit, if y seen in fit was Panel or Hierarchical.

  • Entries are variance forecasts, for var in col index.

A variance forecast for given variable and fh index is a predicted marginal variance for that variable and index, given observed data.

pretrain(y, X=None, fh=None)[source]#

Pre-train forecaster on panel (global) data.

Pre-trains the forecaster by learning shared patterns from multiple time series (panel or hierarchical data). After pretraining, the forecaster can be fine-tuned on a specific series via fit, which preserves the pretrained weights instead of resetting them.

Only forecasters with the capability:pretrain tag set to True implement meaningful pretraining logic. For all other forecasters, pretrain is a no-op that still transitions the state.

If called when the forecaster is already in the "pretrained" or "fitted" state, the internal _pretrain_update method is called instead, enabling incremental pretraining on additional data batches.

State change:

Changes state from "new" to "pretrained".

Writes to self:

  • Sets pretrained model attributes ending in "_". These are inspectable via get_pretrained_params.

  • Sets self.state to "pretrained".

  • Sets self._pretrained_attrs to list of pretrained attribute names (as strings).

Parameters:
ypd.DataFrame with MultiIndex, or other Panel/Hierarchical mtype

Panel or hierarchical time series data to pretrain on. Must contain multiple time series instances (not a single series). For pd-multiindex mtype, the index should have levels (instance, timepoint).

Xpd.DataFrame, optional (default=None)

Exogenous time series, if supported by the forecaster.

fhint, list, np.array, or ForecastingHorizon, optional (default=None)

Forecasting horizon. Not required by all forecasters. For neural network based forecasters, may need to match the network’s output dimension (e.g., pred_len).

Returns:
selfreference to self
Raises:
TypeError

If y is a single Series instead of Panel/Hierarchical data.

TypeError

If the forecaster does not natively support the input data and would require vectorization (not supported for pretraining).

See also

fit

Fit forecaster to a single series (fine-tunes after pretraining).

get_pretrained_params

Retrieve parameters set during pretraining.

state

Current state of the forecaster.

Examples

>>> from sktime.forecasting.dummy_global import DummyGlobalForecaster
>>> from sktime.utils._testing.hierarchical import _make_hierarchical
>>> y_panel = _make_hierarchical(
...     hierarchy_levels=(3,), min_timepoints=12, max_timepoints=12
... )
>>> forecaster = DummyGlobalForecaster()
>>> forecaster.pretrain(y_panel)
DummyGlobalForecaster()
>>> forecaster.state
'pretrained'
reset()[source]#

Reset the object to a clean post-init state.

Results in setting self to the state it had directly after the constructor call, with the same hyper-parameters. Config values set by set_config are also retained.

A reset call deletes any object attributes, except:

  • hyper-parameters = arguments of __init__ written to self, e.g., self.paramname where paramname is an argument of __init__

  • object attributes containing double-underscores, i.e., the string “__”. For instance, an attribute named “__myattr” is retained.

  • config attributes, configs are retained without change. That is, results of get_config before and after reset are equal.

Class and object methods, and class attributes are also unaffected.

Equivalent to clone, with the exception that reset mutates self instead of returning a new object.

After a self.reset() call, self is equal in value and state, to the object obtained after a constructor call``type(self)(**self.get_params(deep=False))``.

Returns:
self

Instance of class reset to a clean post-init state but retaining the current hyper-parameter values.

save(path=None, serialization_format='pickle')[source]#

Save serialized self to bytes-like object or to (.zip) file.

Behaviour:

  • if path is None, returns an in-memory serialized self

  • if path is a file location, stores self at that location as a zip file

saved files are zip files with following contents:

  • _metadata - contains class of self, i.e., type(self)

  • _obj - serialized self. This class uses the default serialization (pickle).

Parameters:
pathNone or file location (str or Path)

if None, self is saved to an in-memory object if file location, self is saved to that file location. If:

  • path=”estimator” then a zip file estimator.zip will be made at cwd.

  • path=”/home/stored/estimator” then a zip file estimator.zip will be

stored in /home/stored/.

serialization_format: str, default = “pickle”

Module to use for serialization. The available options are “pickle” and “cloudpickle”. Note that non-default formats might require installation of other soft dependencies.

Returns:
if path is None - in-memory serialized self
if path is file location - ZipFile with reference to the file
score(y, X=None, fh=None)[source]#

Scores forecast against ground truth, using MAPE (non-symmetric).

Parameters:
ypd.Series, pd.DataFrame, or np.ndarray (1D or 2D)

Time series to score

fhint, list, pd.Index coercible, or ForecastingHorizon, default=None

The forecasting horizon encoding the time stamps to forecast at.

Xpd.DataFrame, or 2D np.array, optional (default=None)

Exogeneous time series to score if self.get_tag(“X-y-must-have-same-index”), X.index must contain y.index

Returns:
scorefloat

MAPE loss of self.predict(fh, X) with respect to y_test.

See also

sktime.performance_metrics.forecasting.mean_absolute_percentage_error
set_config(**config_dict)[source]#

Set config flags to given values.

Parameters:
config_dictdict

Dictionary of config name : config value pairs. Valid configs, values, and their meaning is listed below:

displaystr, “diagram” (default), or “text”

how jupyter kernels display instances of self

  • “diagram” = html box diagram representation

  • “text” = string printout

print_changed_onlybool, default=True

whether printing of self lists only self-parameters that differ from defaults (False), or all parameter names and values (False). Does not nest, i.e., only affects self and not component estimators.

warningsstr, “on” (default), or “off”

whether to raise warnings, affects warnings from sktime only

  • “on” = will raise warnings from sktime

  • “off” = will not raise warnings from sktime

backend:parallelstr, optional, default=”None”

backend to use for parallelization when broadcasting/vectorizing, one of

  • “None”: executes loop sequentially, simple list comprehension

  • “loky”, “multiprocessing” and “threading”: uses joblib.Parallel

  • “joblib”: custom and 3rd party joblib backends, e.g., spark

  • “dask”: uses dask, requires dask package in environment

  • “ray”: uses ray, requires ray package in environment

backend:parallel:paramsdict, optional, default={} (no parameters passed)

additional parameters passed to the parallelization backend as config. Valid keys depend on the value of backend:parallel:

  • “None”: no additional parameters, backend_params is ignored

  • “loky”, “multiprocessing” and “threading”: default joblib backends any valid keys for joblib.Parallel can be passed here, e.g., n_jobs, with the exception of backend which is directly controlled by backend. If n_jobs is not passed, it will default to -1, other parameters will default to joblib defaults.

  • “joblib”: custom and 3rd party joblib backends, e.g., spark. Any valid keys for joblib.Parallel can be passed here, e.g., n_jobs, backend must be passed as a key of backend_params in this case. If n_jobs is not passed, it will default to -1, other parameters will default to joblib defaults.

  • “dask”: any valid keys for dask.compute can be passed, e.g., scheduler

  • “ray”: The following keys can be passed:

    • “ray_remote_args”: dictionary of valid keys for ray.init

    • “shutdown_ray”: bool, default=True; False prevents ray from

      shutting down after parallelization.

    • “logger_name”: str, default=”ray”; name of the logger to use.

    • “mute_warnings”: bool, default=False; if True, suppresses warnings

remember_databool, default=True

whether self._X and self._y are stored in fit, and updated in update. If True, self._X and self._y are stored and updated. If False, self._X and self._y are not stored and updated. This reduces serialization size when using save, but the default update with update_params=True will not refit (only the cutoff advances) and will warn; use UpdateRefitsEvery from sktime.forecasting.stream to pool and refit explicitly.

Returns:
selfreference to self.

Notes

Changes object state, copies configs in config_dict to self._config_dynamic.

set_params(**params)[source]#

Set the parameters of this object.

The method works on simple skbase objects as well as on composite objects. Parameter key strings <component>__<parameter> can be used for composites, i.e., objects that contain other objects, to access <parameter> in the component <component>. The string <parameter>, without <component>__, can also be used if this makes the reference unambiguous, e.g., there are no two parameters of components with the name <parameter>.

Parameters:
**paramsdict

BaseObject parameters, keys must be <component>__<parameter> strings. __ suffixes can alias full strings, if unique among get_params keys.

Returns:
selfreference to self (after parameters have been set)
set_random_state(random_state=None, deep=True, self_policy='copy')[source]#

Set random_state pseudo-random seed parameters for self.

Finds random_state named parameters via self.get_params, and sets them to integers derived from random_state via set_params. These integers are sampled from chain hashing via sample_dependent_seed, and guarantee pseudo-random independence of seeded random generators.

Applies to random_state parameters in self, depending on self_policy, and remaining component objects if and only if deep=True.

Note: calls set_params even if self does not have a random_state, or none of the components have a random_state parameter. Therefore, set_random_state will reset any scikit-base object, even those without a random_state parameter.

Parameters:
random_stateint, RandomState instance or None, default=None

Pseudo-random number generator to control the generation of the random integers. Pass int for reproducible output across multiple function calls.

deepbool, default=True

Whether to set the random state in skbase object valued parameters, i.e., component estimators.

  • If False, will set only self’s random_state parameter, if exists.

  • If True, will set random_state parameters in component objects as well.

self_policystr, one of {“copy”, “keep”, “new”}, default=”copy”
  • “copy” : self.random_state is set to input random_state

  • “keep” : self.random_state is kept as is

  • “new” : self.random_state is set to a new random state,

derived from input random_state, and in general different from it

Returns:
selfreference to self
set_tags(**tag_dict)[source]#

Set instance level tag overrides to given values.

Every scikit-base compatible object has a dictionary of tags. Tags may be used to store metadata about the object, or to control behaviour of the object.

Tags are key-value pairs specific to an instance self, they are static flags that are not changed after construction of the object.

set_tags sets dynamic tag overrides to the values as specified in tag_dict, with keys being the tag name, and dict values being the value to set the tag to.

The set_tags method should be called only in the __init__ method of an object, during construction, or directly after construction via __init__.

Current tag values can be inspected by get_tags or get_tag.

Parameters:
**tag_dictdict

Dictionary of tag name: tag value pairs.

Returns:
Self

Reference to self.

property state[source]#

State of the estimator.

Possible states are:

  • “new”: post-init state

  • “pretrained”: after pretrain has been called, until fit is called

  • “fitted”: after fit has been called

Returns:
str, one of {“new”, “pretrained”, “fitted”}

State of the estimator.

update(y, X=None, update_params=True)[source]#

Update cutoff value and, optionally, fitted parameters.

If no estimator-specific update method has been implemented, default fall-back is as follows:

  • update_params=True: fitting to all observed data so far

  • update_params=False: updates cutoff and remembers data only

State required:

Requires state to be “fitted”, i.e., self.is_fitted=True.

Accesses in self:

  • Fitted model attributes ending in “_”.

  • self.cutoff, self.is_fitted

Writes to self:

  • Updates self.cutoff to latest index seen in y.

  • If update_params=True, updates fitted model attributes ending in “_”.

Parameters:
ytime series in sktime compatible data container format.

Time series with which to update the forecaster.

Individual data formats in sktime are so-called mtype specifications, each mtype implements an abstract scitype.

  • Series scitype = individual time series, vanilla forecasting. pd.DataFrame, pd.Series, or np.ndarray (1D or 2D)

  • Panel scitype = collection of time series, global/panel forecasting. pd.DataFrame with 2-level row MultiIndex (instance, time), 3D np.ndarray (instance, variable, time), list of Series typed pd.DataFrame

  • Hierarchical scitype = hierarchical collection, for hierarchical forecasting. pd.DataFrame with 3 or more level row MultiIndex (hierarchy_1, ..., hierarchy_n, time)

For further details on data format, see glossary on mtype. For usage, see forecasting tutorial examples/01_forecasting.ipynb

Xtime series in sktime compatible format, optional (default=None).

Exogeneous time series to update the model fit with Should be of same scitype (Series, Panel, or Hierarchical) as y. If self.get_tag("X-y-must-have-same-index"), X.index must contain y.index.

update_paramsbool, optional (default=True)

whether model parameters should be updated. If False, only the cutoff is updated, model parameters (e.g., coefficients) are not updated.

Returns:
selfreference to self
update_predict(y, cv=None, X=None, update_params=True, reset_forecaster=True)[source]#

Make predictions and update model iteratively over the test set.

Shorthand to carry out chain of multiple update / predict executions, with data playback based on temporal splitter cv.

Same as the following (if only y, cv are non-default):

  1. self.update(y=cv.split_series(y)[0][0])

  2. remember self.predict() (return later in single batch)

  3. self.update(y=cv.split_series(y)[1][0])

  4. remember self.predict() (return later in single batch)

  5. etc

  6. return all remembered predictions

If no estimator-specific update method has been implemented, default fall-back is as follows:

  • update_params=True: fitting to all observed data so far

  • update_params=False: updates cutoff and remembers data only

State required:

Requires state to be “fitted”, i.e., self.is_fitted=True.

Accesses in self:

  • Fitted model attributes ending in “_”.

  • self.cutoff, self.is_fitted

Writes to self (unless reset_forecaster=True):
  • Updates self.cutoff to latest index seen in y.

  • If update_params=True, updates fitted model attributes ending in “_”.

Does not update state if reset_forecaster=True.

Parameters:
ytime series in sktime compatible data container format.

Time series with which to update the forecaster.

Individual data formats in sktime are so-called mtype specifications, each mtype implements an abstract scitype.

  • Series scitype = individual time series, vanilla forecasting. pd.DataFrame, pd.Series, or np.ndarray (1D or 2D)

  • Panel scitype = collection of time series, global/panel forecasting. pd.DataFrame with 2-level row MultiIndex (instance, time), 3D np.ndarray (instance, variable, time), list of Series typed pd.DataFrame

  • Hierarchical scitype = hierarchical collection, for hierarchical forecasting. pd.DataFrame with 3 or more level row MultiIndex (hierarchy_1, ..., hierarchy_n, time)

For further details on data format, see glossary on mtype. For usage, see forecasting tutorial examples/01_forecasting.ipynb

cvtemporal cross-validation generator inheriting from BaseSplitter, optional

for example, SlidingWindowSplitter or ExpandingWindowSplitter; default = ExpandingWindowSplitter with initial_window=1 and defaults = individual data points in y/X are added and forecast one-by-one, initial_window = 1, step_length = 1 and fh = 1

Xtime series in sktime compatible format, optional (default=None)

Exogeneous time series for updating and forecasting Should be of same scitype (Series, Panel, or Hierarchical) as y in fit. If self.get_tag("X-y-must-have-same-index"), X.index must contain fh index reference.

update_paramsbool, optional (default=True)

whether model parameters should be updated. If False, only the cutoff is updated, model parameters (e.g., coefficients) are not updated.

reset_forecasterbool, optional (default=True)
  • if True, will not change the state of the forecaster, i.e., update/predict sequence is run with a copy, and cutoff, model parameters, data memory of self do not change

  • if False, will update self when the update/predict sequence is run as if update/predict were called directly

Returns:
y_predobject that tabulates point forecasts from multiple split batches

format depends on pairs (cutoff, absolute horizon) forecast overall

  • if collection of absolute horizon points is unique: type is time series in sktime compatible data container format cutoff is suppressed in output has same type as the y that has been passed most recently: Series, Panel, Hierarchical scitype, same format (see above)

  • if collection of absolute horizon points is not unique: type is a pandas DataFrame, with row and col index being time stamps row index corresponds to cutoffs that are predicted from column index corresponds to absolute horizons that are predicted entry is the point prediction of col index predicted from row index entry is nan if no prediction is made at that (cutoff, horizon) pair

update_predict_single(y=None, fh=None, X=None, update_params=True)[source]#

Update model with new data and make forecasts.

This method is useful for updating and making forecasts in a single step.

If no estimator-specific update method has been implemented, default fall-back is first update, then predict.

State required:

Requires state to be “fitted”.

Accesses in self:

Fitted model attributes ending in “_”. self.cutoff, self._is_fitted If update_params=True, model attributes ending in “_”.

Writes to self:

Updates self.cutoff and self._cutoff to last index seen in y. If update_params=True,

updates fitted model attributes ending in “_”.

Parameters:
ytime series in sktime compatible data container format.

Time series with which to update the forecaster.

Individual data formats in sktime are so-called mtype specifications, each mtype implements an abstract scitype.

  • Series scitype = individual time series, vanilla forecasting. pd.DataFrame, pd.Series, or np.ndarray (1D or 2D)

  • Panel scitype = collection of time series, global/panel forecasting. pd.DataFrame with 2-level row MultiIndex (instance, time), 3D np.ndarray (instance, variable, time), list of Series typed pd.DataFrame

  • Hierarchical scitype = hierarchical collection, for hierarchical forecasting. pd.DataFrame with 3 or more level row MultiIndex (hierarchy_1, ..., hierarchy_n, time)

For further details on data format, see glossary on mtype. For usage, see forecasting tutorial examples/01_forecasting.ipynb

fhint, list, pd.Index coercible, or ForecastingHorizon, default=None

The forecasting horizon encoding the time stamps to forecast at. Should not be passed if has already been passed in fit. If has not been passed in fit, must be passed, not optional

Xtime series in sktime compatible format, optional (default=None)

Exogeneous time series for updating and forecasting Should be of same scitype (Series, Panel, or Hierarchical) as y in fit. If self.get_tag("X-y-must-have-same-index"), X.index must contain fh index reference.

update_paramsbool, optional (default=True)

whether model parameters should be updated. If False, only the cutoff is updated, model parameters (e.g., coefficients) are not updated.

Returns:
y_predtime series in sktime compatible data container format

Point forecasts at fh, with same index as fh. y_pred has same type as the y that has been passed most recently: Series, Panel, Hierarchical scitype, same format (see above)