The ClustersSelector class

Objects of this class are used to select optimal cluster expansion models, i.e. the optimal set of clusters, for given training data and clusters pool.

Initialization and methods

class clusterx.clusters_selector.ClustersSelector(basis='trigonometric', selector_type='identity', **selector_opts)

Select optimal cluster expansion models from training data and a clusters pool.

Objects of this class select the optimal set of clusters for the given training data and clusters pool. After initializing a ClustersSelector object, call ClustersSelector.select_clusters() to perform the selection. Different optimality criteria can be used.

Parameters:
basisstr, default “trigonometric”

Cluster basis used during the optimization task. For details and allowed values, see clusterx.correlations.CorrelationsCalculator.

selector_typestr, default “identity”

Selection method:

  • "identity": no selection is performed. The optimal clusters pool is the same as the argument cpool in select_clusters().

  • "subsets_cv": cross-validation optimization of models defined by subsets of the clusters pool.

  • "lasso_cv": cross-validation optimization of the sparsity parameter.

  • "lasso_on_residual": LASSO selection on the residual between a model defined by set0 and the output. The final model contains the union of clusters in set0 and those selected by LASSO on the residual.

  • "OMP": Orthogonal Matching Pursuit.

Deprecated names are described in Notes.

**selector_optsdict

Keyword options passed directly to the constructor. These include general options and options specific to selector_type, described below.

Other Parameters:
fit_interceptbool, default True

General option controlling whether to fit an intercept.

cv_type{“loo”, “5-fold”, “10-fold”}, default “loo”

General cross-validation option. Use "loo" for leave-one-out CV, "5-fold" for 5-fold CV, or "10-fold" for 10-fold CV.

cv_shufflebool, default False

General option. If cv_type is n-fold, this specifies whether to shuffle the data for CV.

cv_random_stateint or None

General option. If cv_shuffle is True, this controls the ordering of the shuffling. If None, the shuffling is different in every call. An integer gives the same shuffling for the same input, allowing reproducible results.

standardizebool, default False

General option. If True, standardize correlations.

sparsity_maxfloat, default 1

Positive maximal sparsity parameter for selector_type="lasso_cv".

sparsity_minfloat, default 0.01

Positive minimal sparsity parameter for selector_type="lasso_cv".

sparsity_stepfloat, optional

Positive sparsity step for selector_type="lasso_cv". If set to 0.0, a logarithmic grid from sparsity_max to sparsity_min is created automatically.

max_iterint

Maximum number of iterations for the LASSO algorithm when selector_type="lasso_cv".

tolfloat

Small positive tolerance of the LASSO solution when selector_type="lasso_cv".

sparsity_scale{“log”, “piece_log”}

Sparsity scale for selector_type="lasso_cv".

clusters_sets{“size”, “combinations”, “size+combinations”}

Subset construction strategy for selector_type="subsets_cv":

  • "size": subpools of increasing size are extracted from the initial pool, and cross-validation selects the optimal subpool.

  • "combinations": all possible combinations of clusters from the pool are considered. This can be very computationally demanding.

  • "size+combinations": a fixed pool of clusters up to the size specified by set0 is always kept. Combinations are searched only for subsets of nclmax clusters.

set0array of two numbers

For selector_type="subsets_cv" and clusters_sets="size+combinations", specifies the size of the fixed pool above which a combinatorial search is performed. The two elements [int, float] give the maximum number of cluster points and the maximum radius of the fixed subpool, respectively. This option also defines the initial model for selector_type="lasso_on_residual", as described above.

nclmaxint

For selector_type="subsets_cv" and clusters_sets="size+combinations", the maximum number of clusters in the combinatorial subsets searched on top of the fixed subpool.

alphafloat, default 0

For subset selection, determines the regularization strength of the scikit-learn ridge estimator. If set to 0, the estimator switches to ordinary least squares.

alphaslist or ndarray of float, default []

For subset selection, determines the list of regularization strengths. If len(alphas) > 1, a scikit-learn RidgeCV estimator is used.

n_nonzero_coefsint, default 1

Number of nonzero parameters for selector_type="OMP".

Notes

The deprecated selector type "lasso" is identical to "lasso_cv"; the deprecated type "linreg" is identical to "subsets_cv". The deprecated keyword option method overrides selector_type.

display_info()

Display in screen information about the optimal model

get_optimal_cpool()

Return optimal ClustersPool object

get_optimal_cpool_array()

Return optimal array of clusters

select_clusters(sset, cpool, prop, comat=None)

Select clusters

Returns a subpool containing the optimal set of clusters.

Parameters:
ssetStructuresSet

The structures set corresponding to the training data.

cpoolClustersPool

the clusters pool from which the optimal model is selected.

propstr

property label (must be in sset) of property for which the optimal set of clusters is to be selected.

comat2D ndarray, default “None”

if the correlation matrix was precalculated, you can give it here