Likelihoods for RegressionModel¶

class darts.utils.likelihood_models.sklearn.GaussianLikelihood(n_outputs, random_state=None)[source]¶

Bases: SKLearnLikelihood

Gaussian distribution [1].

Parameters

n_outputs (int) – The number of predicted outputs per model call. 1 if multi_models=False, otherwise output_chunk_length.
random_state (Optional[int, None]) – Optionally, control the randomness of the sampling.

References

1: https://en.wikipedia.org/wiki/Normal_distribution

Attributes

`num_parameters`	Returns the number of distribution parameters for a single target value.
`parameter_names`	Returns the likelihood parameter names.
`type`	Returns the likelihood type.

Methods

`component_names`(input_series)	Generates names for the parameters of the Likelihood.
`predict`(model, x, num_samples, ...)	Generates sampled or direct likelihood parameter predictions.
`predict_likelihood_parameters`(model_output)	Returns the distribution parameters as a array, extracted from the raw model outputs.
`sample`(model_output)	Samples a prediction from the likelihood distribution and the predicted parameters.

component_names(input_series)¶

Generates names for the parameters of the Likelihood.

Return type: list[str]

property num_parameters: int¶

Returns the number of distribution parameters for a single target value.

Return type: int

property parameter_names: list[str]¶

Returns the likelihood parameter names.

Return type: list[str]

predict(model, x, num_samples, predict_likelihood_parameters, **kwargs)¶

Generates sampled or direct likelihood parameter predictions.

Parameters

model – The Darts RegressionModel.
x (ndarray) – The input feature array passed to the underlying estimator’s predict() method.
num_samples (int) – Number of times a prediction is sampled from the likelihood model / distribution. If 1 and predict_likelihood_parameters=False, returns median / mean predictions.
predict_likelihood_parameters (bool) – If set to True, generates likelihood parameter predictions instead of sampling from the likelihood model / distribution. Only supported with num_samples = 1 and n<=output_chunk_length.
kwargs – Some kwargs passed to the underlying estimator’s predict() method.

Return type

ndarray

predict_likelihood_parameters(model_output)[source]¶

Returns the distribution parameters as a array, extracted from the raw model outputs.

Return type: ndarray

sample(model_output)[source]¶

Samples a prediction from the likelihood distribution and the predicted parameters.

Return type: ndarray

property type: LikelihoodType¶

Returns the likelihood type.

Return type: LikelihoodType

class darts.utils.likelihood_models.sklearn.PoissonLikelihood(n_outputs, random_state=None)[source]¶

Bases: SKLearnLikelihood

Poisson distribution [1].

Parameters

n_outputs (int) – The number of predicted outputs per model call. 1 if multi_models=False, otherwise output_chunk_length.
random_state (Optional[int, None]) – Optionally, control the randomness of the sampling.

References

1: https://en.wikipedia.org/wiki/Poisson_distribution

Attributes

`num_parameters`	Returns the number of distribution parameters for a single target value.
`parameter_names`	Returns the likelihood parameter names.
`type`	Returns the likelihood type.

Methods

`component_names`(input_series)	Generates names for the parameters of the Likelihood.
`predict`(model, x, num_samples, ...)	Generates sampled or direct likelihood parameter predictions.
`predict_likelihood_parameters`(model_output)	Returns the distribution parameters as a array, extracted from the raw model outputs.
`sample`(model_output)	Samples a prediction from the likelihood distribution and the predicted parameters.

component_names(input_series)¶

Generates names for the parameters of the Likelihood.

Return type: list[str]

property num_parameters: int¶

Returns the number of distribution parameters for a single target value.

Return type: int

property parameter_names: list[str]¶

Returns the likelihood parameter names.

Return type: list[str]

predict(model, x, num_samples, predict_likelihood_parameters, **kwargs)¶

Generates sampled or direct likelihood parameter predictions.

Parameters

model – The Darts RegressionModel.
x (ndarray) – The input feature array passed to the underlying estimator’s predict() method.
num_samples (int) – Number of times a prediction is sampled from the likelihood model / distribution. If 1 and predict_likelihood_parameters=False, returns median / mean predictions.
predict_likelihood_parameters (bool) – If set to True, generates likelihood parameter predictions instead of sampling from the likelihood model / distribution. Only supported with num_samples = 1 and n<=output_chunk_length.
kwargs – Some kwargs passed to the underlying estimator’s predict() method.

Return type

ndarray

predict_likelihood_parameters(model_output)[source]¶

Returns the distribution parameters as a array, extracted from the raw model outputs.

Return type: ndarray

sample(model_output)[source]¶

Samples a prediction from the likelihood distribution and the predicted parameters.

Return type: ndarray

property type: LikelihoodType¶

Returns the likelihood type.

Return type: LikelihoodType

class darts.utils.likelihood_models.sklearn.QuantileRegression(n_outputs, random_state=None, quantiles=None)[source]¶

Bases: SKLearnLikelihood

Quantile Regression [1].

Parameters

n_outputs (int) – The number of predicted outputs per model call. 1 if multi_models=False, otherwise output_chunk_length.
random_state (Optional[int, None]) – Optionally, control the randomness of the sampling.
quantiles (Optional[list[float], None]) – A list of quantiles. Default: [0.01, 0.05, 0.1, 0.25, 0.5, 0.75, 0.9, 0.95, 0.99].

References

1: https://en.wikipedia.org/wiki/Quantile_regression

Attributes

`num_parameters`	Returns the number of distribution parameters for a single target value.
`parameter_names`	Returns the likelihood parameter names.
`type`	Returns the likelihood type.

Methods

`component_names`(input_series)	Generates names for the parameters of the Likelihood.
`predict`(model, x, num_samples, ...)	Generates sampled or direct likelihood parameter predictions.
`predict_likelihood_parameters`(model_output)	Returns the distribution parameters as a array, extracted from the raw model outputs.
`sample`(model_output)	Samples a prediction from the likelihood distribution and the predicted parameters.

component_names(input_series)¶

Generates names for the parameters of the Likelihood.

Return type: list[str]

property num_parameters: int¶

Returns the number of distribution parameters for a single target value.

Return type: int

property parameter_names: list[str]¶

Returns the likelihood parameter names.

Return type: list[str]

predict(model, x, num_samples, predict_likelihood_parameters, **kwargs)¶

Generates sampled or direct likelihood parameter predictions.

Parameters

model – The Darts RegressionModel.
x (ndarray) – The input feature array passed to the underlying estimator’s predict() method.
num_samples (int) – Number of times a prediction is sampled from the likelihood model / distribution. If 1 and predict_likelihood_parameters=False, returns median / mean predictions.
predict_likelihood_parameters (bool) – If set to True, generates likelihood parameter predictions instead of sampling from the likelihood model / distribution. Only supported with num_samples = 1 and n<=output_chunk_length.
kwargs – Some kwargs passed to the underlying estimator’s predict() method.

Return type

ndarray

predict_likelihood_parameters(model_output)[source]¶

Returns the distribution parameters as a array, extracted from the raw model outputs.

Return type: ndarray

sample(model_output)[source]¶

Samples a prediction from the likelihood distribution and the predicted parameters.

Return type: ndarray

property type: LikelihoodType¶

Returns the likelihood type.

Return type: LikelihoodType

class darts.utils.likelihood_models.sklearn.SKLearnLikelihood(likelihood_type, parameter_names, n_outputs, random_state=None)[source]¶

Bases: Likelihood, ABC

Base class for sklearn wrapper (e.g. RegressionModel) likelihoods.

Parameters

likelihood_type (LikelihoodType) – A pre-defined LikelihoodType.
parameter_names (list[str]) – The likelihood (distribution) parameter names.
n_outputs (int) – The number of predicted outputs per model call. 1 if multi_models=False, otherwise output_chunk_length.
random_state (Optional[int, None]) – Optionally, control the randomness of the sampling.

Attributes

`num_parameters`	Returns the number of distribution parameters for a single target value.
`parameter_names`	Returns the likelihood parameter names.
`type`	Returns the likelihood type.

Methods

`component_names`(input_series)	Generates names for the parameters of the Likelihood.
`predict`(model, x, num_samples, ...)	Generates sampled or direct likelihood parameter predictions.
`predict_likelihood_parameters`(model_output)	Returns the distribution parameters as a array, extracted from the raw model outputs.
`sample`(model_output)	Samples a prediction from the likelihood distribution and the predicted parameters.

component_names(input_series)¶

Generates names for the parameters of the Likelihood.

Return type: list[str]

property num_parameters: int¶

Returns the number of distribution parameters for a single target value.

Return type: int

property parameter_names: list[str]¶

Returns the likelihood parameter names.

Return type: list[str]

predict(model, x, num_samples, predict_likelihood_parameters, **kwargs)[source]¶

Generates sampled or direct likelihood parameter predictions.

Parameters

model – The Darts RegressionModel.
x (ndarray) – The input feature array passed to the underlying estimator’s predict() method.
num_samples (int) – Number of times a prediction is sampled from the likelihood model / distribution. If 1 and predict_likelihood_parameters=False, returns median / mean predictions.
predict_likelihood_parameters (bool) – If set to True, generates likelihood parameter predictions instead of sampling from the likelihood model / distribution. Only supported with num_samples = 1 and n<=output_chunk_length.
kwargs – Some kwargs passed to the underlying estimator’s predict() method.

Return type

ndarray

abstract predict_likelihood_parameters(model_output)[source]¶

Returns the distribution parameters as a array, extracted from the raw model outputs.

Return type: ndarray

abstract sample(model_output)[source]¶

Samples a prediction from the likelihood distribution and the predicted parameters.

Return type: ndarray

property type: LikelihoodType¶

Returns the likelihood type.

Return type: LikelihoodType

Likelihoods for StatsForecast Models