Module reference¶

Overview¶

The scikit-learn library provides functionality for training linear models and a large number of related tools. The present module provides simplified interfaces for various linear model regression methods. These methods are set up in a way that work out of the box for typical problems in cluster expansion and force constant potential construction, including slight adjustments to scikit-learn default values.

The list of available linear regression method through the trainstation interface is

Ordinary Least-Squares (OLS)
Least-Squares with regularization matrix
Least Absolute Shrinkage and Selection Operator (LASSO)
Adaptive-LASSO
Ridge and Bayesian-ridge
Elasticnet
Recursive Feature Elimination (RFE)
Automatic Relevance Determination Regression (ARDR)
Fitting using Orthogonal Matching Pursuit (OMP)
L1-regularization with split-Bregman

The most commonly used fit methods in the present context are LASSO, automatic relevance determination regression (ARDR), recursive feature elimination with \(\ell_2\)-fitting (RFE-L2) as well as ordinary least-squares optimization (OLS). Below follows a short summary of the main algorithms. More information about the available linear models can be found in the scikit-learn documentation.

Least-squares¶

Ordinary least-squares (OLS) optimization is providing a solution to the linear problem

\[\boldsymbol{A}\boldsymbol{x} = \boldsymbol{y},\]

where \(\boldsymbol{A}\) is the sensing matrix, \(\boldsymbol{y}\) is the vector of target values, and \(\boldsymbol{x}\) is the solution (parameter vector) that one seeks to obtain. The objective is given by

\[\left\Vert\boldsymbol{A}\boldsymbol{x} - \boldsymbol{y}\right\Vert^2_2\]

The OLS method is chosen by setting the fit_method keyword to least-squares.

Least-squares with regularization matrix¶

Similar to OLS, least-squares with regularization matrix optimization solves the linear problem

\[\boldsymbol{A}\boldsymbol{x} = \boldsymbol{y},\]

with the difference that in this case, an explicit regularization matrix \(\boldsymbol{\Lambda}\) is present such that the objective becomes

\[\left(\boldsymbol{y}-\boldsymbol{A}\boldsymbol{x} \right)' \left(\boldsymbol{y}-\boldsymbol{A}\boldsymbol{x} \right) + \boldsymbol{x}' \Lambda \boldsymbol{x}.\]

From a Bayesian perspective, the regularization matrix \(\boldsymbol{\Lambda}\) can be thought of as the inverse of the covariance matrix for the prior probability of the features [MueCed09]. It can be used to scale and couple features based on prior beliefs.

The least-squares with regularization matrix method is chosen by setting the fit_method keyword to least-squares-with-reg-matrix. The regularization matrix is set via the reg_matrix keyword. If no matrix is specified, all elements will be set to zero and the OLS result is recovered. Note that if standardization is used, the regularization matrix should still be designed with respect to the input sensing matrix and not the standardized on.

Parameter	Type	Description	Default
`reg_matrix`	`np.ndarray`	regularization matrix	`None`

LASSO¶

The least absolute shrinkage and selection operator (LASSO) is a method for performing variable selection and regularization in problems in statistics and machine learning. The optimization objective is given by

\[\frac{1}{2 n_\text{samples}} \left\Vert\boldsymbol{A}\boldsymbol{x} - \boldsymbol{y}\right\Vert^2_2 + \alpha \Vert\boldsymbol{x}\Vert_1.\]

While the first term ensures that \(\boldsymbol{x}\) is a solution to the linear problem at hand, the second term introduces regularization and guides the algorithm toward finding sparse solutions, in the spirit of compressive sensing. In general, LASSO is suited for solving strongly underdetermined problems.

The LASSO optimizer is chosen by setting the fit_method keyword to lasso. The \(\alpha\) parameter is set via the alpha keyword. If no value is specified a line scan will be carried out automatically to determine the optimal value.

Parameter	Type	Description	Default
`alpha`	`float`	controls the sparsity of the solution vector	`None`

Automatic relevance determination regression (ARDR)¶

Automatic relevance determination regression (ARDR) is an optimization algorithm provided by scikit-learn that is similar to Bayesian Ridge Regression, which provides a probabilistic model of the regression problem at hand. The method is also known as Sparse Bayesian Learning and Relevance Vector Machine.

The ARDR optimizer is chosen by setting the fit_method keyword to ardr. The threshold lambda parameter, which controls the sparsity of the solution vector, is set via the threshold_lambda keyword (default: 1e4).

Parameter	Type	Description	Default
`threshold_lambda`	`float`	controls the sparsity of the solution vector	`1e4`

split-Bregman¶

The split-Bregman method [GolOsh09] is designed to solve a broad class of \(\ell_1\)-regularized problems. The solution vector \(\boldsymbol{x}\) is given by

\[\boldsymbol{x} = \arg\min_{\boldsymbol{x}, \boldsymbol{d}} \left\Vert\boldsymbol{d}\right\Vert_1 + \frac{1}{2} \left\Vert\boldsymbol{A}\boldsymbol{x} - \boldsymbol{y}\right\Vert^2 + \frac{\lambda}{2} \left\Vert\boldsymbol{d} - \mu \boldsymbol{x} \right\Vert^2,\]

where \(\boldsymbol{d}\) is an auxiliary quantity, while \(\mu\) and \(\lambda\) are hyperparameters that control the sparseness of the solution and the efficiency of the algorithm.

The approach can be optimized by addition of a preconditioning step [ZhoSadAbe19]. This speed-up enables efficient hyperparameter optimization of mu values. By default, the split-bregman fit method will trial a range of \(\mu\) values and choose the optimal based on cross validation.

The split-Bregman implementation supports the following keywords.

Parameter	Type	Description	Default
`mu`	`float`	sparseness parameter	`None`
`lmbda`	`float`	weight of additional L2-norm in split-Bregman	`3`
`n_iters`	`int`	maximal number of split-Bregman iterations	`1000`
`tol`	`float`	convergence criterion of iterative minimization	`1e-4`
`cg_tol`	`float`	convergence criterion of conjugate gradient step	`1e-1`
`cg_n_iters`	`float`	maximal number of conjugate gradient iterations	`None`
`cv_splits`	`int`	number of CV splits for finding optimal mu value	`5`
`iprint`	`int`	how often to print fitting information to stdout	`False`

Recursive feature elimination¶

Recursive feature elimination (RFE) is a feature selection algorithm that obtains the optimal features by carrying out a series of fits, starting with the full set of parameters and then iteratively eliminating the less important ones. RFE needs to be combined with a specific fit method. Since RFE may require many hundreds of single fits its often advisable to use ordinary least-squares as training method, which is the default behavior. The present implementation is based on the implementation of feature selection in scikit-learn.

The RFE optimizer is chosen by setting the fit_method keyword to rfe. The n_features keyword allows one to specify the number of features to select. If this parameter is left unspecified RFE with cross-validation will be used to determine the optimal number of features.

After the optimal number of features has been determined the final model is trained. The fit method for the final fit can be controlled via final_estimator. Here, estimator and final_estimator can be set to any of the fit methods described in this section. For example, estimator='lasso' implies that a LASSO-CV scan is carried out for each fit in the RFE algorithm.

Parameter	Type	Description	Default
`n_features`	`int`	number of features to select	`None`
`step`	`int`	number parameters to eliminate
	`float`	percentage of parameters to eliminate	`0.04`
`cv_splits`	`int`	number of CV splits (90/10) used when optimizing `n_features`	`5`
`estimator`	`str`	fit method to be used in RFE algorithm	`'least-squares'`
`final_estimator`	`str`	fit method to be used in the final fit	= `estimator`
`estimator_kwargs`	`dict`	keyword arguments for fit method defined by `estimator`	`{}`
`final_estimator_kwargs`	`dict`	keyword arguments for fit method defined by `final_estimator`	`{}`

Note

When running on multi-core systems please be mindful of memory consumption. By default all CPUs will be used (n_jobs=-1), which will duplicate data and can require a lot of memory, potentially giving rise to errors. To prevent this behavior you can set the n_jobs parameter explicitly, which is handed over directly to scikit-learn.

Other methods¶

The optimizers furthermore support the ridge method (ridge), the elastic net method (elasticnet) as well as Bayesian ridge regression (bayesian-ridge).

Optimizer¶

class trainstation.Optimizer(fit_data, fit_method='least-squares', standardize=True, train_size=0.9, test_size=None, train_set=None, test_set=None, check_condition=True, seed=42, **kwargs)[source]¶

This optimizer finds a solution to the linear \(\boldsymbol{A}\boldsymbol{x}=\boldsymbol{y}\) problem.

One has to specify either train_size/test_size or train_set/test_set If either train_set or test_set (or both) is specified the fractions will be ignored.

Warning

Repeatedly setting up an Optimizer object and training without changing the seed for the random number generator will yield identical or correlated results, to avoid this please specify a different seed when setting up multiple Optimizer instances.

Parameters

fit_data (tuple(numpy.ndarray, numpy.ndarray)) – the first element of the tuple represents the fit matrix A (N, M array) while the second element represents the vector of target values y (N array); here N (=rows of A, elements of y) equals the number of target values and M (=columns of A) equals the number of parameters
fit_method (str) – method to be used for training; possible choice are “ardr”, “bayesian-ridge”, “elasticnet”, “lasso”, “least-squares”, “omp”, “rfe”, “ridge”, “split-bregman”
standardize (bool) – if True the fit matrix and target values are standardized before fitting, meaning columns in the fit matrix and th target values are rescaled to have a standard deviation of 1.0.
train_size (float or int) – If float represents the fraction of fit_data (rows) to be used for training. If int, represents the absolute number of rows to be used for training.
test_size (float or int) – If float represents the fraction of fit_data (rows) to be used for testing. If int, represents the absolute number of rows to be used for testing.
train_set (tuple or list(int)) – indices of rows of A/y to be used for training
test_set (tuple or list(int)) – indices of rows of A/y to be used for testing
check_condition (bool) – if True the condition number will be checked (this can be sligthly more time consuming for larger matrices)
seed (int) – seed for pseudo random number generator

scatter_data_train¶

target and predicted value for each row in the training set

Type: ScatterData

scatter_data_test¶

target and predicted value for each row in the test set

Type: ScatterData

property AIC: float¶

Akaike information criterion (AIC) for the model

Return type: float

property BIC: float¶

Bayesian information criterion (BIC) for the model

Return type: float

property fit_method: str¶

Fit method

Return type: str

get_contributions(A)¶

Returns the average contribution for each row of A to the predicted values from each element of the parameter vector.

Parameters: A (ndarray) – fit matrix where N (=rows of A, elements of y) equals the number of target values and M (=columns of A) equals the number of parameters
Return type: ndarray

property n_nonzero_parameters: int¶

Number of non-zero parameters

Return type: int

property n_parameters: int¶

Number of parameters (=columns in A matrix)

Return type: int

property n_target_values: int¶

Number of target values (=rows in A matrix)

Return type: int

property parameters: numpy.ndarray¶

Copy of parameter vector

Return type: ndarray

property parameters_norm: float¶

Norm of the parameter vector

Return type: float

property rmse_test: float¶

Root mean squared error for test set

Return type: float

property rmse_train: float¶

Root mean squared error for training set

Return type: float

property seed: int¶

Seed used to initialize pseudo random number generator

Return type: int

property standardize: bool¶

If True standardize the fit matrix before fitting

Return type: bool

property summary: Dict[str, Any]¶

Comprehensive information about the optimizer

Return type: Dict[str, Any]

property test_fraction: float¶

Fraction of rows included in test set

Return type: float

property test_set: List[int]¶

Indices of rows included in the test set

Return type: List[int]

property test_size: int¶

Number of rows included in test set

Return type: int

train()[source]¶

Carries out training.

Return type: None

property train_fraction: float¶

Fraction of rows included in training set

Return type: float

property train_set: List[int]¶

Indices of rows included in the training set

Return type: List[int]

property train_size: int¶

Number of rows included in training set

Return type: int

write_summary(fname)¶

Writes summary dict to file.

Return type: None

EnsembleOptimizer¶

class trainstation.EnsembleOptimizer(fit_data, fit_method='least-squares', standardize=True, ensemble_size=50, train_size=1.0, bootstrap=True, check_condition=True, seed=42, **kwargs)[source]¶

The ensemble optimizer carries out a series of single optimization runs using the Optimizer class in order to solve the linear \(\boldsymbol{A}\boldsymbol{x} = \boldsymbol{y}\) problem. Subsequently, it provides access to various ensemble averaged quantities such as errors and parameters.

Warning

Repeatedly setting up a EnsembleOptimizer and training without changing the seed for the random number generator will yield identical or correlated results, to avoid this please specify a different seed when setting up multiple EnsembleOptimizer instances.

Parameters

fit_data (tuple(numpy.ndarray, numpy.ndarray)) – the first element of the tuple represents the fit matrix A (N, M array) while the second element represents the vector of target values y (N array); here N (=rows of A, elements of y) equals the number of target values and M (=columns of A) equals the number of parameters
fit_method (str) – method to be used for training; possible choice are “ardr”, “bayesian-ridge”, “elasticnet”, “lasso”, “least-squares”, “omp”, “rfe”, “ridge”, “split-bregman”
standardize (bool) – if True the fit matrix and target values are standardized before fitting, meaning columns in the fit matrix and th target values are rescaled to have a standard deviation of 1.0.
ensemble_size (int) – number of fits in the ensemble
train_size (float or int) – if float represents the fraction of fit_data (rows) to be used for training; if int, represents the absolute number of rows to be used for training
bootstrap (bool) – if True sampling will be carried out with replacement
check_condition (bool) – if True the condition number will be checked (this can be sligthly more time consuming for larger matrices)
seed (int) – seed for pseudo random number generator

property bootstrap: bool¶

True if sampling is carried out with replacement

Return type: bool

property ensemble_size: int¶

Number of train rounds

Return type: int

property error_matrix: numpy.ndarray¶

Matrix of fit errors where N is the number of target values and M is the number of fits (i.e., the size of the ensemble)

Return type: ndarray

property fit_method: str¶

Fit method

Return type: str

get_contributions(A)¶

Returns the average contribution for each row of A to the predicted values from each element of the parameter vector.

Parameters: A (ndarray) – fit matrix where N (=rows of A, elements of y) equals the number of target values and M (=columns of A) equals the number of parameters
Return type: ndarray

property n_nonzero_parameters: int¶

Number of non-zero parameters

Return type: int

property n_parameters: int¶

Number of parameters (=columns in A matrix)

Return type: int

property n_target_values: int¶

Number of target values (=rows in A matrix)

Return type: int

property parameters: numpy.ndarray¶

Copy of parameter vector

Return type: ndarray

property parameters_norm: float¶

Norm of the parameter vector

Return type: float

property parameters_splits: List[numpy.ndarray]¶

All parameters vectors in the ensemble

Return type: List[ndarray]

property parameters_std: numpy.ndarray¶

Standard deviation for each parameter

Return type: ndarray

predict(A, return_std=False)[source]¶

Predicts data given an input matrix \(\boldsymbol{A}\), i.e., \(\boldsymbol{A}\boldsymbol{x}\), where \(\boldsymbol{x}\) is the vector of the fitted parameters. The method returns the vector of predicted values and optionally also the vector of standard deviations.

By using all parameter vectors in the ensemble a standard deviation of the prediction can be obtained.

Parameters

A (ndarray) – fit matrix where N (=rows of A, elements of y) equals the number of target values and M (=columns of A) equals the number of parameters
return_std (bool) – whether or not to return the standard deviation of the prediction

Return type

Union[ndarray, Tuple[ndarray, ndarray]]

property rmse_test: float¶

Ensemble average of root mean squared error over test sets

Return type: float

property rmse_test_splits: numpy.ndarray¶

Root mean squared test errors obtained during for each fit in ensemble

Return type: ndarray

property rmse_train: float¶

Ensemble average of root mean squared error over train sets

Return type: float

property rmse_train_splits: numpy.ndarray¶

Root mean squared train errors obtained during for each fit in ensemble

Return type: ndarray

property seed: int¶

Seed used to initialize pseudo random number generator

Return type: int

property standardize: bool¶

If True standardize the fit matrix before fitting

Return type: bool

property summary: Dict[str, Any]¶

Comprehensive information about the optimizer

Return type: Dict[str, Any]

train()[source]¶

Carries out ensemble training and construct the final model by averaging over all models in the ensemble.

Return type: None

property train_size: int¶

Number of rows included in train sets; note that this will be different from the number of unique rows if boostrapping

Return type: int

write_summary(fname)¶

Writes summary dict to file.

Return type: None

CrossValidationEstimator¶

class trainstation.CrossValidationEstimator(fit_data, fit_method='least-squares', standardize=True, validation_method='k-fold', n_splits=10, check_condition=True, seed=42, **kwargs)[source]¶

This class provides an optimizer with cross validation for solving the linear \(\boldsymbol{A}\boldsymbol{x} = \boldsymbol{y}\) problem. Cross-validation (CV) scores are calculated by splitting the available reference data in multiple different ways. It also produces the finalized model (using the full input data) for which the CV score is an estimation of its performance.

Warning

Repeatedly setting up a CrossValidationEstimator and training without changing the seed for the random number generator will yield identical or correlated results, to avoid this please specify a different seed when setting up multiple CrossValidationEstimator instances.

Parameters

fit_data (tuple(numpy.ndarray, numpy.ndarray)) – the first element of the tuple represents the fit matrix A (N, M array) while the second element represents the vector of target values y (N array); here N (=rows of A, elements of y) equals the number of target values and M (=columns of A) equals the number of parameters
fit_method (str) – method to be used for training; possible choice are “ardr”, “bayesian-ridge”, “elasticnet”, “lasso”, “least-squares”, “omp”, “rfe”, “ridge”, “split-bregman”
standardize (bool) – if True the fit matrix and target values are standardized before fitting, meaning columns in the fit matrix and th target values are rescaled to have a standard deviation of 1.0.
validation_method (str) – method to use for cross-validation; possible choices are “shuffle-split”, “k-fold”
n_splits (int) – number of times the fit data set will be split for the cross-validation
check_condition (bool) – if True the condition number will be checked (this can be sligthly more time consuming for larger matrices)
seed (int) – seed for pseudo random number generator

scatter_data_train¶

contains target and predicted values from each individual traininig set in the cross-validation split

Type: ScatterData

scatter_data_validation¶

contains target and predicted values from each individual validation set in the cross-validation split

Type: ScatterData

property AIC: float¶

Akaike information criterion (AIC) for the model

Return type: float

property BIC: float¶

Bayesian information criterion (BIC) for the model

Return type: float

property R2_validation: float¶

Average R2 score for validation sets

Return type: float

property fit_method: str¶

Fit method

Return type: str

get_contributions(A)¶

Returns the average contribution for each row of A to the predicted values from each element of the parameter vector.

Parameters: A (ndarray) – fit matrix where N (=rows of A, elements of y) equals the number of target values and M (=columns of A) equals the number of parameters
Return type: ndarray

property n_nonzero_parameters: int¶

Number of non-zero parameters

Return type: int

property n_nonzero_parameters_splits: numpy.ndarray¶

Number of non-zero parameters for each split

Return type: ndarray

property n_parameters: int¶

Number of parameters (=columns in A matrix)

Return type: int

property n_splits: int¶

Number of splits (folds) used for cross-validation

Return type: int

property n_target_values: int¶

Number of target values (=rows in A matrix)

Return type: int

property parameters: numpy.ndarray¶

Copy of parameter vector

Return type: ndarray

property parameters_norm: float¶

Norm of the parameter vector

Return type: float

property parameters_splits: numpy.ndarray¶

All parameters obtained during cross-validation

Return type: ndarray

property rmse_train: float¶

Average root mean squared training error obtained during cross-validation

Return type: float

property rmse_train_final: float¶

Root mean squared error when using the full set of input data

Return type: float

property rmse_train_splits: numpy.ndarray¶

Root mean squared training errors obtained during cross-validation

Return type: ndarray

property rmse_validation: float¶

Average root mean squared cross-validation error

Return type: float

property rmse_validation_splits: numpy.ndarray¶

Root mean squared validation errors obtained during cross-validation

Return type: ndarray

property seed: int¶

Seed used to initialize pseudo random number generator

Return type: int

property standardize: bool¶

If True standardize the fit matrix before fitting

Return type: bool

property summary: Dict[str, Any]¶

Comprehensive information about the optimizer

Return type: Dict[str, Any]

train()[source]¶

Constructs the final model using all input data available.

Return type: None

validate()[source]¶

Runs validation.

Return type: None

property validation_method: str¶

Validation method name

Return type: str

write_summary(fname)¶

Writes summary dict to file.

Return type: None