pce module¶
Polynomial chaos expansion model.
The polynomial chaos expansion (PCE) model expresses an output variable as a weighted sum of polynomial functions which are orthonormal in the stochastic input space spanned by the random input variables:
where \(\phi_i(x)=\psi_{\tau_1(i),1}(x_1)\times\ldots\times \psi_{\tau_d(i),d}(x_d)\) and \(d\) is the number of input variables.
Enumeration strategy¶
The choice of the function \(\tau=(\tau_1,\ldots,\tau_d)\) is an enumeration strategy and \(\tau_j(i)\) is the polynomial degree of \(\psi_{\tau_j(i),j}\).
Distributions¶
PCE models depend on random input variables and are often used to deal with uncertainty quantification problems.
If \(X_j\) is a Gaussian random variable, \((\psi_{ij})_{i\geq 0}\) is the Legendre basis. If \(X_j\) is a uniform random variable, \((\psi_{ij})_{i\geq 0}\) is the Hermite basis.
When the problem is deterministic, we can still use PCE models under the assumption that the input variables are independent uniform random variables. Then, the orthonormal function basis is the Hermite one.
Degree¶
The degree \(P\) of a PCE model is defined in such a way that \(\max_i \text{degree}(\phi_i)=\sum_{j=1}^d\tau_j(i)=P\).
Estimation¶
The coefficients \((w_1, w_2, ..., w_K)\) and the intercept \(w_0\)
are estimated either by least-squares regression or a quadrature rule.
In the case of least-squares regression,
a sparse strategy can be considered with the LARS algorithm
and in both cases,
the CleaningStrategy
can also remove the non-significant coefficients.
Dependence¶
The PCE model relies on the OpenTURNS class FunctionalChaosAlgorithm
.
- class gemseo.mlearning.regression.pce.CleaningOptions(max_considered_terms=100, most_significant=20, significance_factor=0.0001)[source]¶
Bases:
object
The options of the CleaningStrategy.
- Parameters:
- max_considered_terms: int = 100¶
The maximum number of coefficients of the polynomial basis to be considered.
- class gemseo.mlearning.regression.pce.PCERegressor(data, probability_space, transformer=mappingproxy({}), input_names=None, output_names=None, degree=2, discipline=None, use_quadrature=False, use_lars=False, use_cleaning=False, hyperbolic_parameter=1.0, n_quadrature_points=0, cleaning_options=None)[source]¶
Bases:
BaseMLRegressionAlgo
Polynomial chaos expansion model.
See Also: API documentation of the OpenTURNS class FunctionalChaosAlgorithm.
- Parameters:
data (IODataset | None) – The learning dataset required in the case of the least-squares regression or when
discipline
isNone
in the case of quadrature.probability_space (ParameterSpace) – The set of random input variables defined by
OTDistribution
instances.transformer (TransformerType) –
The strategies to transform the variables. The values are instances of
BaseTransformer
while the keys are the names of either the variables or the groups of variables, e.g."inputs"
or"outputs"
in the case of the regression algorithms. If a group is specified, theBaseTransformer
will be applied to all the variables of this group. IfIDENTITY
, do not transform the variables.By default it is set to {}.
input_names (Iterable[str] | None) – The names of the input variables. If
None
, consider all the input variables of the learning dataset.output_names (Iterable[str] | None) – The names of the output variables. If
None
, consider all the output variables of the learning dataset.degree (int) –
The polynomial degree of the PCE.
By default it is set to 2.
discipline (MDODiscipline | None) – The discipline to be sampled if
use_quadrature
isTrue
anddata
isNone
.use_quadrature (bool) –
Whether to estimate the coefficients of the PCE by a quadrature rule; if so, use the quadrature points stored in
data
or samplediscipline
. otherwise, estimate the coefficients by least-squares regression.By default it is set to False.
use_lars (bool) –
Whether to use the LARS algorithm in the case of the least-squares regression.
By default it is set to False.
use_cleaning (bool) –
Whether to use the CleaningStrategy algorithm. Otherwise, use a fixed truncation strategy (FixedStrategy).
By default it is set to False.
hyperbolic_parameter (float) –
The \(q\)-quasi norm parameter of the hyperbolic and anisotropic enumerate function, defined over the interval \(]0,1]\).
By default it is set to 1.0.
n_quadrature_points (int) –
The total number of quadrature points used by the quadrature strategy to compute the marginal number of points by input dimension when
discipline
is notNone
. If0
, use \((1+P)^d\) points, where \(d\) is the dimension of the input space and \(P\) is the polynomial degree of the PCE.By default it is set to 0.
cleaning_options (CleaningOptions | None) – The options of the CleaningStrategy. If
None
, useDEFAULT_CLEANING_OPTIONS
.
- Raises:
ValueError – When both data and discipline are missing, when both data and discipline are provided, when discipline is provided in the case of least-squares regression, when data is missing in the case of least-squares regression, when the probability space does not contain the distribution of an input variable, when an input variable has a data transformer or when a probability distribution is not an
OTDistribution
.
- DataFormatters¶
alias of
RegressionDataFormatters
- learn(samples=None, fit_transformers=True)¶
Train the machine learning algorithm from the learning dataset.
- load_algo(directory)¶
Load a machine learning algorithm from a directory.
- Parameters:
directory (str | Path) – The path to the directory where the machine learning algorithm is saved.
- Return type:
None
- predict(input_data)¶
Predict output data from input data.
The user can specify these input data either as a NumPy array, e.g.
array([1., 2., 3.])
or as a dictionary, e.g.{'a': array([1.]), 'b': array([2., 3.])}
.If the numpy arrays are of dimension 2, their i-th rows represent the input data of the i-th sample; while if the numpy arrays are of dimension 1, there is a single sample.
The type of the output data and the dimension of the output arrays will be consistent with the type of the input data and the size of the input arrays.
- predict_jacobian(input_data)¶
Predict the Jacobians of the regression model at input_data.
The user can specify these input data either as a NumPy array, e.g.
array([1., 2., 3.])
or as a dictionary, e.g.{'a': array([1.]), 'b': array([2., 3.])}
.If the NumPy arrays are of dimension 2, their i-th rows represent the input data of the i-th sample; while if the NumPy arrays are of dimension 1, there is a single sample.
The type of the output data and the dimension of the output arrays will be consistent with the type of the input data and the size of the input arrays.
- Parameters:
input_data (DataType) – The input data.
- Returns:
The predicted Jacobian data.
- Return type:
NoReturn
- predict_raw(input_data)¶
Predict output data from input data.
- Parameters:
input_data (RealArray) – The input data with shape (n_samples, n_inputs).
- Returns:
The predicted output data with shape (n_samples, n_outputs).
- Return type:
RealArray
- to_pickle(directory=None, path='.', save_learning_set=False)¶
Save the machine learning algorithm.
- Parameters:
directory (str | None) – The name of the directory to save the algorithm.
path (str | Path) –
The path to parent directory where to create the directory.
By default it is set to “.”.
save_learning_set (bool) –
Whether to save the learning set or get rid of it to lighten the saved files.
By default it is set to False.
- Returns:
The path to the directory where the algorithm is saved.
- Return type:
- DEFAULT_TRANSFORMER: DefaultTransformerType = mappingproxy({'inputs': <gemseo.mlearning.transformers.scaler.min_max_scaler.MinMaxScaler object>, 'outputs': <gemseo.mlearning.transformers.scaler.min_max_scaler.MinMaxScaler object>})¶
The default transformer for the input and output data, if any.
- IDENTITY: Final[DefaultTransformerType] = mappingproxy({})¶
A transformer leaving the input and output variables as they are.
- LIBRARY: ClassVar[str] = 'OpenTURNS'¶
The name of the library of the wrapped machine learning algorithm.
- SHORT_ALGO_NAME: ClassVar[str] = 'PCE'¶
The short name of the machine learning algorithm, often an acronym.
Typically used for composite names, e.g.
f"{algo.SHORT_ALGO_NAME}_{dataset.name}"
orf"{algo.SHORT_ALGO_NAME}_{discipline.name}"
.
- algo: Any¶
The interfaced machine learning algorithm.
- property covariance: RealArray¶
The covariance matrix of the PCE model output.
Warning
This statistic is expressed in relation to the transformed output space. You can sample the
predict()
method to estimate it in relation to the original output space if it is different from the transformed output space.
- property first_sobol_indices: list[dict[str, float]]¶
The first-order Sobol’ indices for the different output components.
Warning
These statistics are expressed in relation to the transformed output space. You can use a
SobolAnalysis
to estimate them in relation to the original output space if it is different from the transformed output space.
- property learning_samples_indices: Sequence[int]¶
The indices of the learning samples used for the training.
- property mean: RealArray¶
The mean vector of the PCE model output.
Warning
This statistic is expressed in relation to the transformed output space. You can sample the
predict()
method to estimate it in relation to the original output space if it is different from the transformed output space.
- resampling_results: dict[str, tuple[BaseResampler, list[BaseMLAlgo], list[ndarray] | ndarray]]¶
The resampler class names bound to the resampling results.
A resampling result is formatted as
(resampler, ml_algos, predictions)
whereresampler
is aBaseResampler
,ml_algos
is the list of the associated machine learning algorithms built during the resampling stage andpredictions
are the predictions obtained with the latter.resampling_results
stores only one resampling result per resampler type (e.g.,"CrossValidation"
,"LeaveOneOut"
and"Boostrap"
).
- property second_sobol_indices: list[dict[str, dict[str, float]]]¶
The second-order Sobol’ indices for the different output components.
Warning
These statistics are expressed in relation to the transformed output space. You can use a
SobolAnalysis
to estimate them in relation to the original output space if it is different from the transformed output space.
- property standard_deviation: RealArray¶
The standard deviation vector of the PCE model output.
Warning
This statistic is expressed in relation to the transformed output space. You can sample the
predict()
method to estimate it in relation to the original output space if it is different from the transformed output space.
- property total_sobol_indices: list[dict[str, float]]¶
The total Sobol’ indices for the different output components.
Warning
These statistics are expressed in relation to the transformed output space. You can use a
SobolAnalysis
to estimate them in relation to the original output space if it is different from the transformed output space.
- transformer: dict[str, BaseTransformer]¶
The strategies to transform the variables, if any.
The values are instances of
BaseTransformer
while the keys are the names of either the variables or the groups of variables, e.g. “inputs” or “outputs” in the case of the regression algorithms. If a group is specified, theBaseTransformer
will be applied to all the variables of this group.
- property variance: RealArray¶
The variance vector of the PCE model output.
Warning
This statistic is expressed in relation to the transformed output space. You can sample the
predict()
method to estimate it in relation to the original output space if it is different from the transformed output space.