Skip to content

sklego/deprecated-in-sklego

Rule deprecated-in-sklego
Pack scikit-lego
Severity warning
Fix safe, with xpd check --fix
Library versions scikit-lego 0.7 and later

These names, keywords and attributes warn on scikit-lego and will be removed. Use the replacement that the message names.

folds = KlusterFoldValidation(cluster_method=KMeans(3))
splitter = TimeGapSplit(date_serie=dates, train_duration=timedelta(days=10), valid_duration=timedelta(days=5))
width = capper.n_columns_
folds = ClusterFoldValidation(cluster_method=KMeans(3))
splitter = TimeGapSplit(date_series=dates, train_duration=timedelta(days=10), valid_duration=timedelta(days=5))
width = capper.n_features_in_

The rule reports the names, keywords and fitted attributes that raise a DeprecationWarning:

Old form Write instead
KlusterFoldValidation ClusterFoldValidation
PandasTypeSelector TypeSelector
TimeGapSplit(date_serie=) date_series=
the attributes dim_, num_fit_cols_, n_columns_ and num_cols_ n_features_in_
n_levels_ n_fitted_levels_
coefs_ coef_
penalty="none" on the fairness classifiers penalty=None
the properties allowed_strategies, allowed_methods and allowed_effects no fix

xpd check --fix applies the fix of this rule.

The safe fix writes ClusterFoldValidation for KlusterFoldValidation, TypeSelector for PandasTypeSelector, date_series= for date_serie=, n_features_in_ for dim_, num_fit_cols_, n_columns_ and num_cols_, n_fitted_levels_ for n_levels_, coef_ for coefs_, and penalty=None for penalty="none". allowed_strategies, allowed_methods and allowed_effects have no fix.

Every fix was run before and after on the pack's sample file, and the folds, the selected columns, the attribute values, the coefficients and the predictions were equal.

An old attribute name is reported only on an object that the file builds from a scikit-lego class with that property. The object is a call chain that starts at the class, as in ColumnCapper().fit(X).n_columns_, or a name that the file assigns such a chain to:

Attribute Classes
dim_ RandomRegressor, RandomAdder, DictMapper, OutlierRemover
num_fit_cols_ the two mixture naive Bayes classifiers
n_columns_ ColumnCapper
num_cols_ IntervalEncoder
n_levels_ HierarchicalRegressor, HierarchicalClassifier

coefs_ is reported only in a file that names ProbWeightRegression or DeadZoneRegressor and no MLP, because scikit-learn's MLP models have their own coefs_. The same name on any other object, such as cfg.dim_ on your own class, is left alone, because the rename would raise AttributeError there. An assignment to the name, and the definition of a function or class of the name, are the file's own and are left alone. Each document carries since with the release that began to warn.

xpd prints this message with the issue. A name that starts with $ stands for code from your file.

KlusterFoldValidation warns since scikit-lego 0.8.2 and only calls ClusterFoldValidation. Use ClusterFoldValidation.

Browse the other rules of the scikit-lego pack.