""" The :mod:`sklearn.compose._column_transformer` module implements utilities to work with heterogeneous data and to apply different transformers to different columns. """ # Author: Andreas Mueller # Joris Van den Bossche # License: BSD import warnings from collections import Counter from itertools import chain from numbers import Integral, Real import numpy as np from scipy import sparse from ..base import TransformerMixin, _fit_context, clone from ..pipeline import _fit_transform_one, _name_estimators, _transform_one from ..preprocessing import FunctionTransformer from ..utils import Bunch, _get_column_indices, _safe_indexing from ..utils._estimator_html_repr import _VisualBlock from ..utils._metadata_requests import METHODS from ..utils._param_validation import HasMethods, Hidden, Interval, StrOptions from ..utils._set_output import ( _get_container_adapter, _get_output_config, _safe_set_output, ) from ..utils.metadata_routing import ( MetadataRouter, MethodMapping, _raise_for_params, _routing_enabled, process_routing, ) from ..utils.metaestimators import _BaseComposition from ..utils.parallel import Parallel, delayed from ..utils.validation import ( _check_feature_names_in, _get_feature_names, _is_pandas_df, _num_samples, check_array, check_is_fitted, ) __all__ = ["ColumnTransformer", "make_column_transformer", "make_column_selector"] _ERR_MSG_1DCOLUMN = ( "1D data passed to a transformer that expects 2D data. " "Try to specify the column selection as a list of one " "item instead of a scalar." ) class ColumnTransformer(TransformerMixin, _BaseComposition): """Applies transformers to columns of an array or pandas DataFrame. This estimator allows different columns or column subsets of the input to be transformed separately and the features generated by each transformer will be concatenated to form a single feature space. This is useful for heterogeneous or columnar data, to combine several feature extraction mechanisms or transformations into a single transformer. Read more in the :ref:`User Guide `. .. versionadded:: 0.20 Parameters ---------- transformers : list of tuples List of (name, transformer, columns) tuples specifying the transformer objects to be applied to subsets of the data. name : str Like in Pipeline and FeatureUnion, this allows the transformer and its parameters to be set using ``set_params`` and searched in grid search. transformer : {'drop', 'passthrough'} or estimator Estimator must support :term:`fit` and :term:`transform`. Special-cased strings 'drop' and 'passthrough' are accepted as well, to indicate to drop the columns or to pass them through untransformed, respectively. columns : str, array-like of str, int, array-like of int, \ array-like of bool, slice or callable Indexes the data on its second axis. Integers are interpreted as positional columns, while strings can reference DataFrame columns by name. A scalar string or int should be used where ``transformer`` expects X to be a 1d array-like (vector), otherwise a 2d array will be passed to the transformer. A callable is passed the input data `X` and can return any of the above. To select multiple columns by name or dtype, you can use :obj:`make_column_selector`. remainder : {'drop', 'passthrough'} or estimator, default='drop' By default, only the specified columns in `transformers` are transformed and combined in the output, and the non-specified columns are dropped. (default of ``'drop'``). By specifying ``remainder='passthrough'``, all remaining columns that were not specified in `transformers`, but present in the data passed to `fit` will be automatically passed through. This subset of columns is concatenated with the output of the transformers. For dataframes, extra columns not seen during `fit` will be excluded from the output of `transform`. By setting ``remainder`` to be an estimator, the remaining non-specified columns will use the ``remainder`` estimator. The estimator must support :term:`fit` and :term:`transform`. Note that using this feature requires that the DataFrame columns input at :term:`fit` and :term:`transform` have identical order. sparse_threshold : float, default=0.3 If the output of the different transformers contains sparse matrices, these will be stacked as a sparse matrix if the overall density is lower than this value. Use ``sparse_threshold=0`` to always return dense. When the transformed output consists of all dense data, the stacked result will be dense, and this keyword will be ignored. n_jobs : int, default=None Number of jobs to run in parallel. ``None`` means 1 unless in a :obj:`joblib.parallel_backend` context. ``-1`` means using all processors. See :term:`Glossary ` for more details. transformer_weights : dict, default=None Multiplicative weights for features per transformer. The output of the transformer is multiplied by these weights. Keys are transformer names, values the weights. verbose : bool, default=False If True, the time elapsed while fitting each transformer will be printed as it is completed. verbose_feature_names_out : bool, default=True If True, :meth:`ColumnTransformer.get_feature_names_out` will prefix all feature names with the name of the transformer that generated that feature. If False, :meth:`ColumnTransformer.get_feature_names_out` will not prefix any feature names and will error if feature names are not unique. .. versionadded:: 1.0 Attributes ---------- transformers_ : list The collection of fitted transformers as tuples of (name, fitted_transformer, column). `fitted_transformer` can be an estimator, or `'drop'`; `'passthrough'` is replaced with an equivalent :class:`~sklearn.preprocessing.FunctionTransformer`. In case there were no columns selected, this will be the unfitted transformer. If there are remaining columns, the final element is a tuple of the form: ('remainder', transformer, remaining_columns) corresponding to the ``remainder`` parameter. If there are remaining columns, then ``len(transformers_)==len(transformers)+1``, otherwise ``len(transformers_)==len(transformers)``. named_transformers_ : :class:`~sklearn.utils.Bunch` Read-only attribute to access any transformer by given name. Keys are transformer names and values are the fitted transformer objects. sparse_output_ : bool Boolean flag indicating whether the output of ``transform`` is a sparse matrix or a dense numpy array, which depends on the output of the individual transformers and the `sparse_threshold` keyword. output_indices_ : dict A dictionary from each transformer name to a slice, where the slice corresponds to indices in the transformed output. This is useful to inspect which transformer is responsible for which transformed feature(s). .. versionadded:: 1.0 n_features_in_ : int Number of features seen during :term:`fit`. Only defined if the underlying transformers expose such an attribute when fit. .. versionadded:: 0.24 feature_names_in_ : ndarray of shape (`n_features_in_`,) Names of features seen during :term:`fit`. Defined only when `X` has feature names that are all strings. .. versionadded:: 1.0 See Also -------- make_column_transformer : Convenience function for combining the outputs of multiple transformer objects applied to column subsets of the original feature space. make_column_selector : Convenience function for selecting columns based on datatype or the columns name with a regex pattern. Notes ----- The order of the columns in the transformed feature matrix follows the order of how the columns are specified in the `transformers` list. Columns of the original feature matrix that are not specified are dropped from the resulting transformed feature matrix, unless specified in the `passthrough` keyword. Those columns specified with `passthrough` are added at the right to the output of the transformers. Examples -------- >>> import numpy as np >>> from sklearn.compose import ColumnTransformer >>> from sklearn.preprocessing import Normalizer >>> ct = ColumnTransformer( ... [("norm1", Normalizer(norm='l1'), [0, 1]), ... ("norm2", Normalizer(norm='l1'), slice(2, 4))]) >>> X = np.array([[0., 1., 2., 2.], ... [1., 1., 0., 1.]]) >>> # Normalizer scales each row of X to unit norm. A separate scaling >>> # is applied for the two first and two last elements of each >>> # row independently. >>> ct.fit_transform(X) array([[0. , 1. , 0.5, 0.5], [0.5, 0.5, 0. , 1. ]]) :class:`ColumnTransformer` can be configured with a transformer that requires a 1d array by setting the column to a string: >>> from sklearn.feature_extraction.text import CountVectorizer >>> from sklearn.preprocessing import MinMaxScaler >>> import pandas as pd # doctest: +SKIP >>> X = pd.DataFrame({ ... "documents": ["First item", "second one here", "Is this the last?"], ... "width": [3, 4, 5], ... }) # doctest: +SKIP >>> # "documents" is a string which configures ColumnTransformer to >>> # pass the documents column as a 1d array to the CountVectorizer >>> ct = ColumnTransformer( ... [("text_preprocess", CountVectorizer(), "documents"), ... ("num_preprocess", MinMaxScaler(), ["width"])]) >>> X_trans = ct.fit_transform(X) # doctest: +SKIP For a more detailed example of usage, see :ref:`sphx_glr_auto_examples_compose_plot_column_transformer_mixed_types.py`. """ _required_parameters = ["transformers"] _parameter_constraints: dict = { "transformers": [list, Hidden(tuple)], "remainder": [ StrOptions({"drop", "passthrough"}), HasMethods(["fit", "transform"]), HasMethods(["fit_transform", "transform"]), ], "sparse_threshold": [Interval(Real, 0, 1, closed="both")], "n_jobs": [Integral, None], "transformer_weights": [dict, None], "verbose": ["verbose"], "verbose_feature_names_out": ["boolean"], } def __init__( self, transformers, *, remainder="drop", sparse_threshold=0.3, n_jobs=None, transformer_weights=None, verbose=False, verbose_feature_names_out=True, ): self.transformers = transformers self.remainder = remainder self.sparse_threshold = sparse_threshold self.n_jobs = n_jobs self.transformer_weights = transformer_weights self.verbose = verbose self.verbose_feature_names_out = verbose_feature_names_out @property def _transformers(self): """ Internal list of transformer only containing the name and transformers, dropping the columns. DO NOT USE: This is for the implementation of get_params via BaseComposition._get_params which expects lists of tuples of len 2. To iterate through the transformers, use ``self._iter`` instead. """ try: return [(name, trans) for name, trans, _ in self.transformers] except (TypeError, ValueError): return self.transformers @_transformers.setter def _transformers(self, value): """DO NOT USE: This is for the implementation of set_params via BaseComposition._get_params which gives lists of tuples of len 2. """ try: self.transformers = [ (name, trans, col) for ((name, trans), (_, _, col)) in zip(value, self.transformers) ] except (TypeError, ValueError): self.transformers = value def set_output(self, *, transform=None): """Set the output container when `"transform"` and `"fit_transform"` are called. Calling `set_output` will set the output of all estimators in `transformers` and `transformers_`. Parameters ---------- transform : {"default", "pandas"}, default=None Configure output of `transform` and `fit_transform`. - `"default"`: Default output format of a transformer - `"pandas"`: DataFrame output - `"polars"`: Polars output - `None`: Transform configuration is unchanged .. versionadded:: 1.4 `"polars"` option was added. Returns ------- self : estimator instance Estimator instance. """ super().set_output(transform=transform) transformers = ( trans for _, trans, _ in chain( self.transformers, getattr(self, "transformers_", []) ) if trans not in {"passthrough", "drop"} ) for trans in transformers: _safe_set_output(trans, transform=transform) if self.remainder not in {"passthrough", "drop"}: _safe_set_output(self.remainder, transform=transform) return self def get_params(self, deep=True): """Get parameters for this estimator. Returns the parameters given in the constructor as well as the estimators contained within the `transformers` of the `ColumnTransformer`. Parameters ---------- deep : bool, default=True If True, will return the parameters for this estimator and contained subobjects that are estimators. Returns ------- params : dict Parameter names mapped to their values. """ return self._get_params("_transformers", deep=deep) def set_params(self, **kwargs): """Set the parameters of this estimator. Valid parameter keys can be listed with ``get_params()``. Note that you can directly set the parameters of the estimators contained in `transformers` of `ColumnTransformer`. Parameters ---------- **kwargs : dict Estimator parameters. Returns ------- self : ColumnTransformer This estimator. """ self._set_params("_transformers", **kwargs) return self def _iter(self, fitted, column_as_labels, skip_drop, skip_empty_columns): """ Generate (name, trans, column, weight) tuples. Parameters ---------- fitted : bool If True, use the fitted transformers (``self.transformers_``) to iterate through transformers, else use the transformers passed by the user (``self.transformers``). column_as_labels : bool If True, columns are returned as string labels. If False, columns are returned as they were given by the user. This can only be True if the ``ColumnTransformer`` is already fitted. skip_drop : bool If True, 'drop' transformers are filtered out. skip_empty_columns : bool If True, transformers with empty selected columns are filtered out. Yields ------ A generator of tuples containing: - name : the name of the transformer - transformer : the transformer object - columns : the columns for that transformer - weight : the weight of the transformer """ if fitted: transformers = self.transformers_ else: # interleave the validated column specifiers transformers = [ (name, trans, column) for (name, trans, _), column in zip(self.transformers, self._columns) ] # add transformer tuple for remainder if self._remainder[2]: transformers = chain(transformers, [self._remainder]) get_weight = (self.transformer_weights or {}).get for name, trans, columns in transformers: if skip_drop and trans == "drop": continue if skip_empty_columns and _is_empty_column_selection(columns): continue if column_as_labels: # Convert all columns to using their string labels columns_is_scalar = np.isscalar(columns) indices = self._transformer_to_input_indices[name] columns = self.feature_names_in_[indices] if columns_is_scalar: # selection is done with one dimension columns = columns[0] yield (name, trans, columns, get_weight(name)) def _validate_transformers(self): """Validate names of transformers and the transformers themselves. This checks whether given transformers have the required methods, i.e. `fit` or `fit_transform` and `transform` implemented. """ if not self.transformers: return names, transformers, _ = zip(*self.transformers) # validate names self._validate_names(names) # validate estimators for t in transformers: if t in ("drop", "passthrough"): continue if not (hasattr(t, "fit") or hasattr(t, "fit_transform")) or not hasattr( t, "transform" ): # Used to validate the transformers in the `transformers` list raise TypeError( "All estimators should implement fit and " "transform, or can be 'drop' or 'passthrough' " "specifiers. '%s' (type %s) doesn't." % (t, type(t)) ) def _validate_column_callables(self, X): """ Converts callable column specifications. This stores a dictionary of the form `{step_name: column_indices}` and calls the `columns` on `X` if `columns` is a callable for a given transformer. The results are then stored in `self._transformer_to_input_indices`. """ all_columns = [] transformer_to_input_indices = {} for name, _, columns in self.transformers: if callable(columns): columns = columns(X) all_columns.append(columns) transformer_to_input_indices[name] = _get_column_indices(X, columns) self._columns = all_columns self._transformer_to_input_indices = transformer_to_input_indices def _validate_remainder(self, X): """ Validates ``remainder`` and defines ``_remainder`` targeting the remaining columns. """ cols = set(chain(*self._transformer_to_input_indices.values())) remaining = sorted(set(range(self.n_features_in_)) - cols) self._remainder = ("remainder", self.remainder, remaining) self._transformer_to_input_indices["remainder"] = remaining @property def named_transformers_(self): """Access the fitted transformer by name. Read-only attribute to access any transformer by given name. Keys are transformer names and values are the fitted transformer objects. """ # Use Bunch object to improve autocomplete return Bunch(**{name: trans for name, trans, _ in self.transformers_}) def _get_feature_name_out_for_transformer(self, name, trans, feature_names_in): """Gets feature names of transformer. Used in conjunction with self._iter(fitted=True) in get_feature_names_out. """ column_indices = self._transformer_to_input_indices[name] names = feature_names_in[column_indices] # An actual transformer if not hasattr(trans, "get_feature_names_out"): raise AttributeError( f"Transformer {name} (type {type(trans).__name__}) does " "not provide get_feature_names_out." ) return trans.get_feature_names_out(names) def get_feature_names_out(self, input_features=None): """Get output feature names for transformation. Parameters ---------- input_features : array-like of str or None, default=None Input features. - If `input_features` is `None`, then `feature_names_in_` is used as feature names in. If `feature_names_in_` is not defined, then the following input feature names are generated: `["x0", "x1", ..., "x(n_features_in_ - 1)"]`. - If `input_features` is an array-like, then `input_features` must match `feature_names_in_` if `feature_names_in_` is defined. Returns ------- feature_names_out : ndarray of str objects Transformed feature names. """ check_is_fitted(self) input_features = _check_feature_names_in(self, input_features) # List of tuples (name, feature_names_out) transformer_with_feature_names_out = [] for name, trans, *_ in self._iter( fitted=True, column_as_labels=False, skip_empty_columns=True, skip_drop=True, ): feature_names_out = self._get_feature_name_out_for_transformer( name, trans, input_features ) if feature_names_out is None: continue transformer_with_feature_names_out.append((name, feature_names_out)) if not transformer_with_feature_names_out: # No feature names return np.array([], dtype=object) return self._add_prefix_for_feature_names_out( transformer_with_feature_names_out ) def _add_prefix_for_feature_names_out(self, transformer_with_feature_names_out): """Add prefix for feature names out that includes the transformer names. Parameters ---------- transformer_with_feature_names_out : list of tuples of (str, array-like of str) The tuple consistent of the transformer's name and its feature names out. Returns ------- feature_names_out : ndarray of shape (n_features,), dtype=str Transformed feature names. """ if self.verbose_feature_names_out: # Prefix the feature names out with the transformers name names = list( chain.from_iterable( (f"{name}__{i}" for i in feature_names_out) for name, feature_names_out in transformer_with_feature_names_out ) ) return np.asarray(names, dtype=object) # verbose_feature_names_out is False # Check that names are all unique without a prefix feature_names_count = Counter( chain.from_iterable(s for _, s in transformer_with_feature_names_out) ) top_6_overlap = [ name for name, count in feature_names_count.most_common(6) if count > 1 ] top_6_overlap.sort() if top_6_overlap: if len(top_6_overlap) == 6: # There are more than 5 overlapping names, we only show the 5 # of the feature names names_repr = str(top_6_overlap[:5])[:-1] + ", ...]" else: names_repr = str(top_6_overlap) raise ValueError( f"Output feature names: {names_repr} are not unique. Please set " "verbose_feature_names_out=True to add prefixes to feature names" ) return np.concatenate( [name for _, name in transformer_with_feature_names_out], ) def _update_fitted_transformers(self, transformers): """Set self.transformers_ from given transformers. Parameters ---------- transformers : list of estimators The fitted estimators as the output of `self._call_func_on_transformers(func=_fit_transform_one, ...)`. That function doesn't include 'drop' or transformers for which no column is selected. 'drop' is kept as is, and for the no-column transformers the unfitted transformer is put in `self.transformers_`. """ # transformers are fitted; excludes 'drop' cases fitted_transformers = iter(transformers) transformers_ = [] for name, old, column, _ in self._iter( fitted=False, column_as_labels=False, skip_drop=False, skip_empty_columns=False, ): if old == "drop": trans = "drop" elif _is_empty_column_selection(column): trans = old else: trans = next(fitted_transformers) transformers_.append((name, trans, column)) # sanity check that transformers is exhausted assert not list(fitted_transformers) self.transformers_ = transformers_ def _validate_output(self, result): """ Ensure that the output of each transformer is 2D. Otherwise hstack can raise an error or produce incorrect results. """ names = [ name for name, _, _, _ in self._iter( fitted=True, column_as_labels=False, skip_drop=True, skip_empty_columns=True, ) ] for Xs, name in zip(result, names): if not getattr(Xs, "ndim", 0) == 2 and not hasattr(Xs, "__dataframe__"): raise ValueError( "The output of the '{0}' transformer should be 2D (numpy array, " "scipy sparse array, dataframe).".format(name) ) if _get_output_config("transform", self)["dense"] == "pandas": return try: import pandas as pd except ImportError: return for Xs, name in zip(result, names): if not _is_pandas_df(Xs): continue for col_name, dtype in Xs.dtypes.to_dict().items(): if getattr(dtype, "na_value", None) is not pd.NA: continue if pd.NA not in Xs[col_name].values: continue class_name = self.__class__.__name__ # TODO(1.6): replace warning with ValueError warnings.warn( ( f"The output of the '{name}' transformer for column" f" '{col_name}' has dtype {dtype} and uses pandas.NA to" " represent null values. Storing this output in a numpy array" " can cause errors in downstream scikit-learn estimators, and" " inefficiencies. Starting with scikit-learn version 1.6, this" " will raise a ValueError. To avoid this problem you can (i)" " store the output in a pandas DataFrame by using" f" {class_name}.set_output(transform='pandas') or (ii) modify" f" the input data or the '{name}' transformer to avoid the" " presence of pandas.NA (for example by using" " pandas.DataFrame.astype)." ), FutureWarning, ) def _record_output_indices(self, Xs): """ Record which transformer produced which column. """ idx = 0 self.output_indices_ = {} for transformer_idx, (name, _, _, _) in enumerate( self._iter( fitted=True, column_as_labels=False, skip_drop=True, skip_empty_columns=True, ) ): n_columns = Xs[transformer_idx].shape[1] self.output_indices_[name] = slice(idx, idx + n_columns) idx += n_columns # `_iter` only generates transformers that have a non empty # selection. Here we set empty slices for transformers that # generate no output, which are safe for indexing all_names = [t[0] for t in self.transformers] + ["remainder"] for name in all_names: if name not in self.output_indices_: self.output_indices_[name] = slice(0, 0) def _log_message(self, name, idx, total): if not self.verbose: return None return "(%d of %d) Processing %s" % (idx, total, name) def _call_func_on_transformers(self, X, y, func, column_as_labels, routed_params): """ Private function to fit and/or transform on demand. Parameters ---------- X : {array-like, dataframe} of shape (n_samples, n_features) The data to be used in fit and/or transform. y : array-like of shape (n_samples,) Targets. func : callable Function to call, which can be _fit_transform_one or _transform_one. column_as_labels : bool Used to iterate through transformers. If True, columns are returned as strings. If False, columns are returned as they were given by the user. Can be True only if the ``ColumnTransformer`` is already fitted. routed_params : dict The routed parameters as the output from ``process_routing``. Returns ------- Return value (transformers and/or transformed X data) depends on the passed function. """ if func is _fit_transform_one: fitted = False else: # func is _transform_one fitted = True transformers = list( self._iter( fitted=fitted, column_as_labels=column_as_labels, skip_drop=True, skip_empty_columns=True, ) ) try: jobs = [] for idx, (name, trans, column, weight) in enumerate(transformers, start=1): if func is _fit_transform_one: if trans == "passthrough": output_config = _get_output_config("transform", self) trans = FunctionTransformer( accept_sparse=True, check_inverse=False, feature_names_out="one-to-one", ).set_output(transform=output_config["dense"]) extra_args = dict( message_clsname="ColumnTransformer", message=self._log_message(name, idx, len(transformers)), ) else: # func is _transform_one extra_args = {} jobs.append( delayed(func)( transformer=clone(trans) if not fitted else trans, X=_safe_indexing(X, column, axis=1), y=y, weight=weight, **extra_args, params=routed_params[name], ) ) return Parallel(n_jobs=self.n_jobs)(jobs) except ValueError as e: if "Expected 2D array, got 1D array instead" in str(e): raise ValueError(_ERR_MSG_1DCOLUMN) from e else: raise def fit(self, X, y=None, **params): """Fit all transformers using X. Parameters ---------- X : {array-like, dataframe} of shape (n_samples, n_features) Input data, of which specified subsets are used to fit the transformers. y : array-like of shape (n_samples,...), default=None Targets for supervised learning. **params : dict, default=None Parameters to be passed to the underlying transformers' ``fit`` and ``transform`` methods. You can only pass this if metadata routing is enabled, which you can enable using ``sklearn.set_config(enable_metadata_routing=True)``. .. versionadded:: 1.4 Returns ------- self : ColumnTransformer This estimator. """ _raise_for_params(params, self, "fit") # we use fit_transform to make sure to set sparse_output_ (for which we # need the transformed data) to have consistent output type in predict self.fit_transform(X, y=y, **params) return self @_fit_context( # estimators in ColumnTransformer.transformers are not validated yet prefer_skip_nested_validation=False ) def fit_transform(self, X, y=None, **params): """Fit all transformers, transform the data and concatenate results. Parameters ---------- X : {array-like, dataframe} of shape (n_samples, n_features) Input data, of which specified subsets are used to fit the transformers. y : array-like of shape (n_samples,), default=None Targets for supervised learning. **params : dict, default=None Parameters to be passed to the underlying transformers' ``fit`` and ``transform`` methods. You can only pass this if metadata routing is enabled, which you can enable using ``sklearn.set_config(enable_metadata_routing=True)``. .. versionadded:: 1.4 Returns ------- X_t : {array-like, sparse matrix} of \ shape (n_samples, sum_n_components) Horizontally stacked results of transformers. sum_n_components is the sum of n_components (output dimension) over transformers. If any result is a sparse matrix, everything will be converted to sparse matrices. """ _raise_for_params(params, self, "fit_transform") self._check_feature_names(X, reset=True) X = _check_X(X) # set n_features_in_ attribute self._check_n_features(X, reset=True) self._validate_transformers() n_samples = _num_samples(X) self._validate_column_callables(X) self._validate_remainder(X) if _routing_enabled(): routed_params = process_routing(self, "fit_transform", **params) else: routed_params = self._get_empty_routing() result = self._call_func_on_transformers( X, y, _fit_transform_one, column_as_labels=False, routed_params=routed_params, ) if not result: self._update_fitted_transformers([]) # All transformers are None return np.zeros((n_samples, 0)) Xs, transformers = zip(*result) # determine if concatenated output will be sparse or not if any(sparse.issparse(X) for X in Xs): nnz = sum(X.nnz if sparse.issparse(X) else X.size for X in Xs) total = sum( X.shape[0] * X.shape[1] if sparse.issparse(X) else X.size for X in Xs ) density = nnz / total self.sparse_output_ = density < self.sparse_threshold else: self.sparse_output_ = False self._update_fitted_transformers(transformers) self._validate_output(Xs) self._record_output_indices(Xs) return self._hstack(list(Xs), n_samples=n_samples) def transform(self, X, **params): """Transform X separately by each transformer, concatenate results. Parameters ---------- X : {array-like, dataframe} of shape (n_samples, n_features) The data to be transformed by subset. **params : dict, default=None Parameters to be passed to the underlying transformers' ``transform`` method. You can only pass this if metadata routing is enabled, which you can enable using ``sklearn.set_config(enable_metadata_routing=True)``. .. versionadded:: 1.4 Returns ------- X_t : {array-like, sparse matrix} of \ shape (n_samples, sum_n_components) Horizontally stacked results of transformers. sum_n_components is the sum of n_components (output dimension) over transformers. If any result is a sparse matrix, everything will be converted to sparse matrices. """ _raise_for_params(params, self, "transform") check_is_fitted(self) X = _check_X(X) # If ColumnTransformer is fit using a dataframe, and now a dataframe is # passed to be transformed, we select columns by name instead. This # enables the user to pass X at transform time with extra columns which # were not present in fit time, and the order of the columns doesn't # matter. fit_dataframe_and_transform_dataframe = hasattr(self, "feature_names_in_") and ( _is_pandas_df(X) or hasattr(X, "__dataframe__") ) n_samples = _num_samples(X) column_names = _get_feature_names(X) if fit_dataframe_and_transform_dataframe: named_transformers = self.named_transformers_ # check that all names seen in fit are in transform, unless # they were dropped non_dropped_indices = [ ind for name, ind in self._transformer_to_input_indices.items() if name in named_transformers and named_transformers[name] != "drop" ] all_indices = set(chain(*non_dropped_indices)) all_names = set(self.feature_names_in_[ind] for ind in all_indices) diff = all_names - set(column_names) if diff: raise ValueError(f"columns are missing: {diff}") else: # ndarray was used for fitting or transforming, thus we only # check that n_features_in_ is consistent self._check_n_features(X, reset=False) if _routing_enabled(): routed_params = process_routing(self, "transform", **params) else: routed_params = self._get_empty_routing() Xs = self._call_func_on_transformers( X, None, _transform_one, column_as_labels=fit_dataframe_and_transform_dataframe, routed_params=routed_params, ) self._validate_output(Xs) if not Xs: # All transformers are None return np.zeros((n_samples, 0)) return self._hstack(list(Xs), n_samples=n_samples) def _hstack(self, Xs, *, n_samples): """Stacks Xs horizontally. This allows subclasses to control the stacking behavior, while reusing everything else from ColumnTransformer. Parameters ---------- Xs : list of {array-like, sparse matrix, dataframe} The container to concatenate. n_samples : int The number of samples in the input data to checking the transformation consistency. """ if self.sparse_output_: try: # since all columns should be numeric before stacking them # in a sparse matrix, `check_array` is used for the # dtype conversion if necessary. converted_Xs = [ check_array(X, accept_sparse=True, force_all_finite=False) for X in Xs ] except ValueError as e: raise ValueError( "For a sparse output, all columns should " "be a numeric or convertible to a numeric." ) from e return sparse.hstack(converted_Xs).tocsr() else: Xs = [f.toarray() if sparse.issparse(f) else f for f in Xs] adapter = _get_container_adapter("transform", self) if adapter and all(adapter.is_supported_container(X) for X in Xs): # rename before stacking as it avoids to error on temporary duplicated # columns transformer_names = [ t[0] for t in self._iter( fitted=True, column_as_labels=False, skip_drop=True, skip_empty_columns=True, ) ] feature_names_outs = [X.columns for X in Xs if X.shape[1] != 0] if self.verbose_feature_names_out: # `_add_prefix_for_feature_names_out` takes care about raising # an error if there are duplicated columns. feature_names_outs = self._add_prefix_for_feature_names_out( list(zip(transformer_names, feature_names_outs)) ) else: # check for duplicated columns and raise if any feature_names_outs = list(chain.from_iterable(feature_names_outs)) feature_names_count = Counter(feature_names_outs) if any(count > 1 for count in feature_names_count.values()): duplicated_feature_names = sorted( name for name, count in feature_names_count.items() if count > 1 ) err_msg = ( "Duplicated feature names found before concatenating the" " outputs of the transformers:" f" {duplicated_feature_names}.\n" ) for transformer_name, X in zip(transformer_names, Xs): if X.shape[1] == 0: continue dup_cols_in_transformer = sorted( set(X.columns).intersection(duplicated_feature_names) ) if len(dup_cols_in_transformer): err_msg += ( f"Transformer {transformer_name} has conflicting " f"columns names: {dup_cols_in_transformer}.\n" ) raise ValueError( err_msg + "Either make sure that the transformers named above " "do not generate columns with conflicting names or set " "verbose_feature_names_out=True to automatically " "prefix to the output feature names with the name " "of the transformer to prevent any conflicting " "names." ) names_idx = 0 for X in Xs: if X.shape[1] == 0: continue names_out = feature_names_outs[names_idx : names_idx + X.shape[1]] adapter.rename_columns(X, names_out) names_idx += X.shape[1] output = adapter.hstack(Xs) output_samples = output.shape[0] if output_samples != n_samples: raise ValueError( "Concatenating DataFrames from the transformer's output lead to" " an inconsistent number of samples. The output may have Pandas" " Indexes that do not match, or that transformers are returning" " number of samples which are not the same as the number input" " samples." ) return output return np.hstack(Xs) def _sk_visual_block_(self): if isinstance(self.remainder, str) and self.remainder == "drop": transformers = self.transformers elif hasattr(self, "_remainder"): remainder_columns = self._remainder[2] if ( hasattr(self, "feature_names_in_") and remainder_columns and not all(isinstance(col, str) for col in remainder_columns) ): remainder_columns = self.feature_names_in_[remainder_columns].tolist() transformers = chain( self.transformers, [("remainder", self.remainder, remainder_columns)] ) else: transformers = chain(self.transformers, [("remainder", self.remainder, "")]) names, transformers, name_details = zip(*transformers) return _VisualBlock( "parallel", transformers, names=names, name_details=name_details ) def _get_empty_routing(self): """Return empty routing. Used while routing can be disabled. TODO: Remove when ``set_config(enable_metadata_routing=False)`` is no more an option. """ return Bunch( **{ name: Bunch(**{method: {} for method in METHODS}) for name, step, _, _ in self._iter( fitted=False, column_as_labels=False, skip_drop=True, skip_empty_columns=True, ) } ) def get_metadata_routing(self): """Get metadata routing of this object. Please check :ref:`User Guide ` on how the routing mechanism works. .. versionadded:: 1.4 Returns ------- routing : MetadataRouter A :class:`~sklearn.utils.metadata_routing.MetadataRouter` encapsulating routing information. """ router = MetadataRouter(owner=self.__class__.__name__) # Here we don't care about which columns are used for which # transformers, and whether or not a transformer is used at all, which # might happen if no columns are selected for that transformer. We # request all metadata requested by all transformers. transformers = chain(self.transformers, [("remainder", self.remainder, None)]) for name, step, _ in transformers: method_mapping = MethodMapping() if hasattr(step, "fit_transform"): ( method_mapping.add(caller="fit", callee="fit_transform").add( caller="fit_transform", callee="fit_transform" ) ) else: ( method_mapping.add(caller="fit", callee="fit") .add(caller="fit", callee="transform") .add(caller="fit_transform", callee="fit") .add(caller="fit_transform", callee="transform") ) method_mapping.add(caller="transform", callee="transform") router.add(method_mapping=method_mapping, **{name: step}) return router def _check_X(X): """Use check_array only when necessary, e.g. on lists and other non-array-likes.""" if hasattr(X, "__array__") or hasattr(X, "__dataframe__") or sparse.issparse(X): return X return check_array(X, force_all_finite="allow-nan", dtype=object) def _is_empty_column_selection(column): """ Return True if the column selection is empty (empty list or all-False boolean array). """ if hasattr(column, "dtype") and np.issubdtype(column.dtype, np.bool_): return not column.any() elif hasattr(column, "__len__"): return ( len(column) == 0 or all(isinstance(col, bool) for col in column) and not any(column) ) else: return False def _get_transformer_list(estimators): """ Construct (name, trans, column) tuples from list """ transformers, columns = zip(*estimators) names, _ = zip(*_name_estimators(transformers)) transformer_list = list(zip(names, transformers, columns)) return transformer_list # This function is not validated using validate_params because # it's just a factory for ColumnTransformer. def make_column_transformer( *transformers, remainder="drop", sparse_threshold=0.3, n_jobs=None, verbose=False, verbose_feature_names_out=True, ): """Construct a ColumnTransformer from the given transformers. This is a shorthand for the ColumnTransformer constructor; it does not require, and does not permit, naming the transformers. Instead, they will be given names automatically based on their types. It also does not allow weighting with ``transformer_weights``. Read more in the :ref:`User Guide `. Parameters ---------- *transformers : tuples Tuples of the form (transformer, columns) specifying the transformer objects to be applied to subsets of the data. transformer : {'drop', 'passthrough'} or estimator Estimator must support :term:`fit` and :term:`transform`. Special-cased strings 'drop' and 'passthrough' are accepted as well, to indicate to drop the columns or to pass them through untransformed, respectively. columns : str, array-like of str, int, array-like of int, slice, \ array-like of bool or callable Indexes the data on its second axis. Integers are interpreted as positional columns, while strings can reference DataFrame columns by name. A scalar string or int should be used where ``transformer`` expects X to be a 1d array-like (vector), otherwise a 2d array will be passed to the transformer. A callable is passed the input data `X` and can return any of the above. To select multiple columns by name or dtype, you can use :obj:`make_column_selector`. remainder : {'drop', 'passthrough'} or estimator, default='drop' By default, only the specified columns in `transformers` are transformed and combined in the output, and the non-specified columns are dropped. (default of ``'drop'``). By specifying ``remainder='passthrough'``, all remaining columns that were not specified in `transformers` will be automatically passed through. This subset of columns is concatenated with the output of the transformers. By setting ``remainder`` to be an estimator, the remaining non-specified columns will use the ``remainder`` estimator. The estimator must support :term:`fit` and :term:`transform`. sparse_threshold : float, default=0.3 If the transformed output consists of a mix of sparse and dense data, it will be stacked as a sparse matrix if the density is lower than this value. Use ``sparse_threshold=0`` to always return dense. When the transformed output consists of all sparse or all dense data, the stacked result will be sparse or dense, respectively, and this keyword will be ignored. n_jobs : int, default=None Number of jobs to run in parallel. ``None`` means 1 unless in a :obj:`joblib.parallel_backend` context. ``-1`` means using all processors. See :term:`Glossary ` for more details. verbose : bool, default=False If True, the time elapsed while fitting each transformer will be printed as it is completed. verbose_feature_names_out : bool, default=True If True, :meth:`ColumnTransformer.get_feature_names_out` will prefix all feature names with the name of the transformer that generated that feature. If False, :meth:`ColumnTransformer.get_feature_names_out` will not prefix any feature names and will error if feature names are not unique. .. versionadded:: 1.0 Returns ------- ct : ColumnTransformer Returns a :class:`ColumnTransformer` object. See Also -------- ColumnTransformer : Class that allows combining the outputs of multiple transformer objects used on column subsets of the data into a single feature space. Examples -------- >>> from sklearn.preprocessing import StandardScaler, OneHotEncoder >>> from sklearn.compose import make_column_transformer >>> make_column_transformer( ... (StandardScaler(), ['numerical_column']), ... (OneHotEncoder(), ['categorical_column'])) ColumnTransformer(transformers=[('standardscaler', StandardScaler(...), ['numerical_column']), ('onehotencoder', OneHotEncoder(...), ['categorical_column'])]) """ # transformer_weights keyword is not passed through because the user # would need to know the automatically generated names of the transformers transformer_list = _get_transformer_list(transformers) return ColumnTransformer( transformer_list, n_jobs=n_jobs, remainder=remainder, sparse_threshold=sparse_threshold, verbose=verbose, verbose_feature_names_out=verbose_feature_names_out, ) class make_column_selector: """Create a callable to select columns to be used with :class:`ColumnTransformer`. :func:`make_column_selector` can select columns based on datatype or the columns name with a regex. When using multiple selection criteria, **all** criteria must match for a column to be selected. For an example of how to use :func:`make_column_selector` within a :class:`ColumnTransformer` to select columns based on data type (i.e. `dtype`), refer to :ref:`sphx_glr_auto_examples_compose_plot_column_transformer_mixed_types.py`. Parameters ---------- pattern : str, default=None Name of columns containing this regex pattern will be included. If None, column selection will not be selected based on pattern. dtype_include : column dtype or list of column dtypes, default=None A selection of dtypes to include. For more details, see :meth:`pandas.DataFrame.select_dtypes`. dtype_exclude : column dtype or list of column dtypes, default=None A selection of dtypes to exclude. For more details, see :meth:`pandas.DataFrame.select_dtypes`. Returns ------- selector : callable Callable for column selection to be used by a :class:`ColumnTransformer`. See Also -------- ColumnTransformer : Class that allows combining the outputs of multiple transformer objects used on column subsets of the data into a single feature space. Examples -------- >>> from sklearn.preprocessing import StandardScaler, OneHotEncoder >>> from sklearn.compose import make_column_transformer >>> from sklearn.compose import make_column_selector >>> import numpy as np >>> import pandas as pd # doctest: +SKIP >>> X = pd.DataFrame({'city': ['London', 'London', 'Paris', 'Sallisaw'], ... 'rating': [5, 3, 4, 5]}) # doctest: +SKIP >>> ct = make_column_transformer( ... (StandardScaler(), ... make_column_selector(dtype_include=np.number)), # rating ... (OneHotEncoder(), ... make_column_selector(dtype_include=object))) # city >>> ct.fit_transform(X) # doctest: +SKIP array([[ 0.90453403, 1. , 0. , 0. ], [-1.50755672, 1. , 0. , 0. ], [-0.30151134, 0. , 1. , 0. ], [ 0.90453403, 0. , 0. , 1. ]]) """ def __init__(self, pattern=None, *, dtype_include=None, dtype_exclude=None): self.pattern = pattern self.dtype_include = dtype_include self.dtype_exclude = dtype_exclude def __call__(self, df): """Callable for column selection to be used by a :class:`ColumnTransformer`. Parameters ---------- df : dataframe of shape (n_features, n_samples) DataFrame to select columns from. """ if not hasattr(df, "iloc"): raise ValueError( "make_column_selector can only be applied to pandas dataframes" ) df_row = df.iloc[:1] if self.dtype_include is not None or self.dtype_exclude is not None: df_row = df_row.select_dtypes( include=self.dtype_include, exclude=self.dtype_exclude ) cols = df_row.columns if self.pattern is not None: cols = cols[cols.str.contains(self.pattern, regex=True)] return cols.tolist()