# Authors: Ronan Perry, Sambit Panda, Haoyin Xu
# Adopted from: https://github.com/neurodata/honest-forests
import threading
from numbers import Integral
from warnings import catch_warnings, simplefilter
import numpy as np
from joblib import Parallel, delayed
from sklearn.base import _fit_context, clone
from sklearn.ensemble._base import _partition_estimators, _set_random_states
from sklearn.utils import compute_sample_weight
from sklearn.utils._param_validation import Interval, RealNotInt
from sklearn.utils.validation import check_is_fitted
from .._lib.sklearn.ensemble._forest import ForestClassifier
from ..tree import HonestTreeClassifier
from ._extensions import ForestClassifierMixin, _generate_sample_indices
def _parallel_build_trees(
tree,
bootstrap,
X,
y,
sample_weight,
tree_idx,
n_trees,
verbose=0,
class_weight=None,
n_samples_bootstrap=None,
missing_values_in_feature_mask=None,
classes=None,
):
"""
Private function used to fit a single tree in parallel.
XXX: this is copied over from scikit-learn and modified to allow sampling with
and without replacement given ``bootstrap``.
"""
if verbose > 1:
print("building tree %d of %d" % (tree_idx + 1, n_trees))
n_samples = X.shape[0]
if sample_weight is None:
curr_sample_weight = np.ones((n_samples,), dtype=np.float64)
else:
curr_sample_weight = sample_weight.copy()
indices = _generate_sample_indices(tree.random_state, n_samples, n_samples_bootstrap, bootstrap)
sample_counts = np.bincount(indices, minlength=n_samples)
curr_sample_weight *= sample_counts
if class_weight == "subsample":
with catch_warnings():
simplefilter("ignore", DeprecationWarning)
curr_sample_weight *= compute_sample_weight("auto", y, indices=indices)
elif class_weight == "balanced_subsample":
curr_sample_weight *= compute_sample_weight("balanced", y, indices=indices)
tree._fit(
X,
y,
sample_weight=curr_sample_weight,
check_input=False,
missing_values_in_feature_mask=missing_values_in_feature_mask,
classes=classes,
)
return tree
[docs]
class HonestForestClassifier(ForestClassifier, ForestClassifierMixin):
"""
A forest classifier with honest leaf estimates.
Parameters
----------
n_estimators : int, default=100
The number of trees in the forest.
criterion : {"gini", "entropy"}, default="gini"
The function to measure the quality of a split. Supported criteria are
"gini" for the Gini impurity and "entropy" for the information gain.
Note: this parameter is tree-specific.
splitter : {"best", "random"}, default="best"
The strategy used to choose the split at each node. Supported
strategies are "best" to choose the best split and "random" to choose
the best random split.
max_depth : int, default=None
The maximum depth of the tree. If None, then nodes are expanded until
all leaves are pure or until all leaves contain less than
min_samples_split samples.
min_samples_split : int or float, default=2
The minimum number of samples required to split an internal node:
- If int, then consider `min_samples_split` as the minimum number.
- If float, then `min_samples_split` is a fraction and
`ceil(min_samples_split * n_samples)` are the minimum
number of samples for each split.
min_samples_leaf : int or float, default=1
The minimum number of samples required to be at a leaf node.
A split point at any depth will only be considered if it leaves at
least ``min_samples_leaf`` training samples in each of the left and
right branches. This may have the effect of smoothing the model,
especially in regression.
- If int, then consider `min_samples_leaf` as the minimum number.
- If float, then `min_samples_leaf` is a fraction and
`ceil(min_samples_leaf * n_samples)` are the minimum
number of samples for each node.
min_weight_fraction_leaf : float, default=0.0
The minimum weighted fraction of the sum total of weights (of all
the input samples) required to be at a leaf node. Samples have
equal weight when sample_weight is not provided.
max_features : {"sqrt", "log2", None}, int or float, default="sqrt"
The number of features to consider when looking for the best split:
- If int, then consider `max_features` features at each split.
- If float, then `max_features` is a fraction and
`round(max_features * n_features)` features are considered at each
split.
- If "auto", then `max_features=sqrt(n_features)`.
- If "sqrt", then `max_features=sqrt(n_features)`.
- If "log2", then `max_features=log2(n_features)`.
- If None, then `max_features=n_features`.
Note: the search for a split does not stop until at least one
valid partition of the node samples is found, even if it requires to
effectively inspect more than ``max_features`` features.
max_leaf_nodes : int, default=None
Grow trees with ``max_leaf_nodes`` in best-first fashion.
Best nodes are defined as relative reduction in impurity.
If None then unlimited number of leaf nodes.
min_impurity_decrease : float, default=0.0
A node will be split if this split induces a decrease of the impurity
greater than or equal to this value.
The weighted impurity decrease equation is the following::
N_t / N * (impurity - N_t_R / N_t * right_impurity
- N_t_L / N_t * left_impurity)
where ``N`` is the total number of samples, ``N_t`` is the number of
samples at the current node, ``N_t_L`` is the number of samples in the
left child, and ``N_t_R`` is the number of samples in the right child.
``N``, ``N_t``, ``N_t_R`` and ``N_t_L`` all refer to the weighted sum,
if ``sample_weight`` is passed.
bootstrap : bool, default=False
Whether bootstrap samples are used when building trees. If False, the
whole dataset is used to build each tree.
When bootstrap is True, each tree bootstrap samples the dataset, and then
the unique indices are split in half, where one half is used to train
the structure of the tree and one half is used to train the leaves of the tree.
The remaining sample indices are considered "out of bag".
oob_score : bool, default=False
Whether to use out-of-bag samples to estimate the generalization score.
Only available if bootstrap=True.
n_jobs : int, default=None
The number of jobs to run in parallel. :meth:`fit`, :meth:`predict`,
:meth:`decision_path` and :meth:`apply` are all parallelized over the
trees. ``None`` means 1 unless in a `joblib.parallel_backend`
context. ``-1`` means using all processors. See :term:`Glossary
<n_jobs>` for more details.
random_state : int, RandomState instance or None, default=None
Controls both the randomness of the bootstrapping of the samples used
when building trees (if ``bootstrap=True``) and the sampling of the
features to consider when looking for the best split at each node
(if ``max_features < n_features``).
See :term:`Glossary <random_state>` for details.
verbose : int, default=0
Controls the verbosity when fitting and predicting.
warm_start : bool, default=False
When set to ``True``, reuse the solution of the previous call to fit
and add more estimators to the ensemble, otherwise, just fit a whole
new forest. See :term:`the Glossary <warm_start>`.
class_weight : {"balanced", "balanced_subsample"}, dict or list of dicts, \
default=None
Weights associated with classes in the form ``{class_label: weight}``.
If not given, all classes are supposed to have weight one. For
multi-output problems, a list of dicts can be provided in the same
order as the columns of y.
Note that for multioutput (including multilabel) weights should be
defined for each class of every column in its own dict. For example,
for four-class multilabel classification weights should be
[{0: 1, 1: 1}, {0: 1, 1: 5}, {0: 1, 1: 1}, {0: 1, 1: 1}] instead of
[{1:1}, {2:5}, {3:1}, {4:1}].
The "balanced" mode uses the values of y to automatically adjust
weights inversely proportional to class frequencies in the input data
as ``n_samples / (n_classes * np.bincount(y))``
The "balanced_subsample" mode is the same as "balanced" except that
weights are computed based on the bootstrap sample for every tree
grown.
For multi-output, the weights of each column of y will be multiplied.
Note that these weights will be multiplied with sample_weight (passed
through the fit method) if sample_weight is specified.
ccp_alpha : non-negative float, default=0.0
Complexity parameter used for Minimal Cost-Complexity Pruning. The
subtree with the largest cost complexity that is smaller than
``ccp_alpha`` will be chosen. By default, no pruning is performed. See
:ref:`minimal_cost_complexity_pruning` for details.
max_samples : int or float, default=None
If bootstrap is True, the number of samples to draw from X
to train each base tree estimator with replacement.
If bootstrap is False, then this will subsample
the dataset without replacement.
- If None (default), then draw `X.shape[0]` samples.
- If int, then draw `max_samples` samples.
- If float, then draw `max_samples * X.shape[0]` samples.
honest_prior : {"ignore", "uniform", "empirical"}, default="empirical"
Method for dealing with empty leaves during evaluation of a test
sample. If "ignore", the tree is ignored. If "uniform", the prior tree
posterior is 1/(number of classes). If "empirical", the prior tree
posterior is the relative class frequency in the voting subsample.
If all trees are ignored, the empirical estimate is returned.
honest_fraction : float, default=0.5
Fraction of training samples used for estimates in the trees. The
remaining samples will be used to learn the tree structure. A larger
fraction creates shallower trees with lower variance estimates.
tree_estimator : object, default=None
Instantiated tree of type BaseDecisionTree from sktree.
If None, then sklearn's DecisionTreeClassifier with default parameters will
be used. Note that none of the parameters in ``tree_estimator`` need
to be set. The parameters of the ``tree_estimator`` can be set using
the ``tree_estimator_params`` keyword argument.
stratify : bool
Whether or not to stratify sample when considering structure and leaf indices.
By default False.
**tree_estimator_params : dict
Parameters to pass to the underlying base tree estimators.
These must be parameters for ``tree_estimator``.
Attributes
----------
estimator : sktree.tree.HonestTreeClassifier
The child estimator template used to create the collection of fitted
sub-estimators.
estimators_ : list of sktree.tree.HonestTreeClassifier
The collection of fitted sub-estimators.
classes_ : ndarray of shape (n_classes,) or a list of such arrays
The classes labels (single output problem), or a list of arrays of
class labels (multi-output problem).
n_classes_ : int or list
The number of classes (single output problem), or a list containing the
number of classes for each output (multi-output problem).
n_features_in_ : int
Number of features seen during :term:`fit`.
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.
n_outputs_ : int
The number of outputs when ``fit`` is performed.
feature_importances_ : ndarray of shape (n_features,)
The impurity-based feature importances.
The higher, the more important the feature.
The importance of a feature is computed as the (normalized)
total reduction of the criterion brought by that feature. It is also
known as the Gini importance.
Warning: impurity-based feature importances can be misleading for
high cardinality features (many unique values). See
:func:`sklearn.inspection.permutation_importance` as an alternative.
oob_score_ : float
Score of the training dataset obtained using an out-of-bag estimate.
This attribute exists only when ``oob_score`` is True.
oob_decision_function_ : ndarray of shape (n_samples, n_classes) or \
(n_samples, n_classes, n_outputs)
Decision function computed with out-of-bag estimate on the training
set. If n_estimators is small it might be possible that a data point
was never left out during the bootstrap. In this case,
`oob_decision_function_` might contain NaN. This attribute exists
only when ``oob_score`` is True.
honest_decision_function_ : ndarray of shape (n_samples, n_classes) or \
(n_samples, n_classes, n_outputs)
Decision function computed on each sample, including only the trees
for which it was in the honest subsample. It is possible that a sample
is never in the honest subset in which case `honest_decision_function_`
might contain NaN.
structure_indices_ : list of lists, shape=(n_estimators, n_structure)
Indices of training samples used to learn the structure.
honest_indices_ : list of lists, shape=(n_estimators, n_honest)
Indices of training samples used to learn leaf estimates.
oob_samples_ : list of lists, shape=(n_estimators, n_samples_bootstrap)
The indices of training samples that are "out-of-bag". Only used
if ``bootstrap=True``.
Notes
-----
The default values for the parameters controlling the size of the trees
(e.g. ``max_depth``, ``min_samples_leaf``, etc.) lead to fully grown and
unpruned trees which can potentially be very large on some data sets. To
reduce memory consumption, the complexity and size of the trees should be
controlled by setting those parameter values.
The features are always randomly permuted at each split. Therefore,
the best found split may vary, even with the same training data,
``max_features=n_features`` and ``bootstrap=False``, if the improvement
of the criterion is identical for several splits enumerated during the
search of the best split. To obtain a deterministic behaviour during
fitting, ``random_state`` has to be fixed.
Honesty is a feature of trees that enables unbiased estimates of confidence
intervals. The default implementation here is using double sampling to
implement honesty. The amount of samples used for learning split nodes vs
leaf nodes is controlled by the ``honest_fraction`` parameter. In order to
enforce honesty, but also enable the tree to have access to all y labels,
we set sample_weight to 0 for a random subset of samples. This results in
inefficiency when building trees using a greedy splitter as we still sort
over all values of X. We recommend using propensity trees if you are
computing causal effects.
This forest classifier is a "meta-estimator" because any tree model can
be used in the classification process, while enabling honesty separates
the data used for split and leaf nodes.
References
----------
.. [1] L. Breiman, "Random Forests", Machine Learning, 45(1), 5-32, 2001.
.. [2] S. Athey, J. Tibshirani, and S. Wager. "Generalized
Random Forests", Annals of Statistics, 2019.
Examples
--------
>>> from honest_forests.estimators import HonestForestClassifier
>>> from sklearn.datasets import make_classification
>>> X, y = make_classification(n_samples=1000, n_features=4,
... n_informative=2, n_redundant=0,
... random_state=0, shuffle=False)
>>> clf = HonestForestClassifier(
>>> max_depth=2,
>>> random_state=0,
>>> tree_estimator=ObliqueDecisionTreeClassifier())
>>> clf.fit(X, y)
HonestForestClassifier(...)
>>> print(clf.predict([[0, 0, 0, 0]]))
[1]
"""
_parameter_constraints: dict = {
**ForestClassifier._parameter_constraints,
}
_parameter_constraints.pop("max_samples")
_parameter_constraints["max_samples"] = [
None,
Interval(RealNotInt, 0.0, None, closed="right"),
Interval(Integral, 1, None, closed="left"),
]
def __init__(
self,
n_estimators=100,
criterion="gini",
splitter="best",
max_depth=None,
min_samples_split=2,
min_samples_leaf=1,
min_weight_fraction_leaf=0.0,
max_features="sqrt",
max_leaf_nodes=None,
min_impurity_decrease=0.0,
bootstrap=False,
oob_score=False,
n_jobs=None,
random_state=None,
verbose=0,
warm_start=False,
class_weight=None,
ccp_alpha=0.0,
max_samples=None,
honest_prior="empirical",
honest_fraction=0.5,
tree_estimator=None,
stratify=False,
**tree_estimator_params,
):
super().__init__(
estimator=HonestTreeClassifier(),
n_estimators=n_estimators,
estimator_params=(
"criterion",
"splitter",
"max_depth",
"min_samples_split",
"min_samples_leaf",
"min_weight_fraction_leaf",
"max_features",
"max_leaf_nodes",
"min_impurity_decrease",
"random_state",
"ccp_alpha",
"tree_estimator",
"honest_fraction",
"honest_prior",
"stratify",
),
bootstrap=bootstrap,
oob_score=oob_score,
n_jobs=n_jobs,
random_state=random_state,
verbose=verbose,
warm_start=warm_start,
class_weight=class_weight,
max_samples=max_samples,
)
self.criterion = criterion
self.splitter = splitter
self.max_depth = max_depth
self.min_samples_split = min_samples_split
self.min_samples_leaf = min_samples_leaf
self.min_weight_fraction_leaf = min_weight_fraction_leaf
self.max_features = max_features
self.max_leaf_nodes = max_leaf_nodes
self.min_impurity_decrease = min_impurity_decrease
self.ccp_alpha = ccp_alpha
self.honest_fraction = honest_fraction
self.honest_prior = honest_prior
self.tree_estimator = tree_estimator
self.stratify = stratify
self._tree_estimator_params = tree_estimator_params
[docs]
@_fit_context(prefer_skip_nested_validation=True)
def fit(self, X, y, sample_weight=None, classes=None, **fit_params):
"""
Build a forest of trees from the training set (X, y).
Parameters
----------
X : {array-like, sparse matrix} of shape (n_samples, n_features)
The training input samples. Internally, its dtype will be converted
to ``dtype=np.float32``. If a sparse matrix is provided, it will be
converted into a sparse ``csc_matrix``.
y : array-like of shape (n_samples,) or (n_samples, n_outputs)
The target values (class labels in classification, real numbers in
regression).
sample_weight : array-like of shape (n_samples,), default=None
Sample weights. If None, then samples are equally weighted. Splits
that would create child nodes with net zero or negative weight are
ignored while searching for a split in each node. In the case of
classification, splits are also ignored if they would result in any
single class carrying a negative weight in either child node.
classes : array-like of shape (n_classes,), default=None
List of all the classes that can possibly appear in the y vector.
**fit_params : dict
Parameters to pass to the underlying base tree estimators.
Only available if `enable_metadata_routing=True`,
which can be set by using
``sklearn.set_config(enable_metadata_routing=True)``.
See :ref:`Metadata Routing User Guide <metadata_routing>` for
more details.
Returns
-------
self : object
Fitted estimator.
"""
super().fit(X, y, sample_weight=sample_weight, classes=classes, **fit_params)
# Inherit attributes from the tree estimator
self._inherit_estimator_attributes()
# Compute honest decision function
self.honest_decision_function_ = self._predict_proba(
X, indices=self.honest_indices_, impute_missing=np.nan
)
return self
def _make_estimator(self, append=True, random_state=None):
"""Make and configure a copy of the `estimator_` attribute.
Warning: This method should be used to properly instantiate new
sub-estimators.
"""
estimator = clone(self.estimator_)
estimator.set_params(**{p: getattr(self, p) for p in self.estimator_params})
# XXX: This is the only change compared to scikit-learn's make_estimator
# additionally set the tree_estimator_params
estimator._tree_estimator_params = self._tree_estimator_params
if random_state is not None:
_set_random_states(estimator, random_state)
if append:
self.estimators_.append(estimator)
return estimator
def _inherit_estimator_attributes(self):
"""Initialize necessary attributes from the provided tree estimator"""
if hasattr(self.tree_estimator, "_inheritable_fitted_attribute"):
for attr in self.tree_estimator._inheritable_fitted_attribute:
setattr(self, attr, getattr(self.estimators_[0], attr))
[docs]
def predict_proba(self, X):
"""
Predict class probabilities for X.
The predicted class probabilities of an input sample are computed as
the mean predicted class probabilities of the trees in the forest.
The class probability of a single tree is the fraction of samples of
the same class in a leaf.
Parameters
----------
X : {array-like, sparse matrix} of shape (n_samples, n_features)
The input samples. Internally, its dtype will be converted to
``dtype=np.float32``. If a sparse matrix is provided, it will be
converted into a sparse ``csr_matrix``.
Returns
-------
p : ndarray of shape (n_samples, n_classes), or a list of such arrays
The class probabilities of the input samples. The order of the
classes corresponds to that in the attribute :term:`classes_`.
"""
return self._predict_proba(X)
def _predict_proba(self, X, indices=None, impute_missing=None):
"""predict_proba helper class"""
check_is_fitted(self)
X = self._validate_X_predict(X)
n_jobs, _, _ = _partition_estimators(self.n_estimators, self.n_jobs)
# avoid storing the output of every tree estimator by summing them here
posteriors = [
np.zeros((X.shape[0], j), dtype=np.float64) for j in np.atleast_1d(self.n_classes_)
]
if indices is None:
indices = [None] * self.n_estimators
lock = threading.Lock()
Parallel(n_jobs=n_jobs, verbose=self.verbose, require="sharedmem")(
delayed(_accumulate_prediction)(tree.predict_proba, X, posteriors, lock, idx)
for tree, idx in zip(self.estimators_, indices)
)
# Normalize to unit length, due to prior weighting
posteriors = np.array(posteriors)
zero_mask = posteriors.sum(2) == 0
posteriors[~zero_mask] /= posteriors[~zero_mask].sum(1, keepdims=True)
if impute_missing is None:
pass
else:
posteriors[zero_mask] = impute_missing
# preserve shape of multi-outputs
if self.n_outputs_ > 1:
posteriors = [post for post in posteriors]
if len(posteriors) == 1:
return posteriors[0]
else:
return posteriors
@property
def structure_indices_(self):
"""The indices used to learn the structure of the trees."""
check_is_fitted(self)
return [tree.structure_indices_ for tree in self.estimators_]
@property
def honest_indices_(self):
"""The indices used to fit the leaf nodes."""
check_is_fitted(self)
return [tree.honest_indices_ for tree in self.estimators_]
@property
def oob_samples_(self):
"""The sample indices that are out-of-bag.
Only utilized if ``bootstrap=True``, otherwise, all samples are "in-bag".
"""
if self.bootstrap is False and (
self._n_samples_bootstrap is None or self._n_samples_bootstrap == self._n_samples
):
raise RuntimeError(
"Cannot extract out-of-bag samples when bootstrap is False and "
"n_samples == n_samples_bootstrap"
)
check_is_fitted(self)
oob_samples = []
possible_indices = np.arange(self._n_samples)
for structure_idx, honest_idx in zip(self.structure_indices_, self.honest_indices_):
_oob_samples = np.setdiff1d(
possible_indices, np.concatenate((structure_idx, honest_idx))
)
oob_samples.append(_oob_samples)
return oob_samples
def _more_tags(self):
return {"multioutput": False}
[docs]
def decision_path(self, X):
"""
Return the decision path in the forest.
.. versionadded:: 0.18
Parameters
----------
X : {array-like, sparse matrix} of shape (n_samples, n_features)
The input samples. Internally, its dtype will be converted to
``dtype=np.float32``. If a sparse matrix is provided, it will be
converted into a sparse ``csr_matrix``.
Returns
-------
indicator : sparse matrix of shape (n_samples, n_nodes)
Return a node indicator matrix where non zero elements indicates
that the samples goes through the nodes. The matrix is of CSR
format.
n_nodes_ptr : ndarray of shape (n_estimators + 1,)
The columns from indicator[n_nodes_ptr[i]:n_nodes_ptr[i+1]]
gives the indicator value for the i-th estimator.
"""
return self.estimator_.decision_path(X)
[docs]
def predict_quantiles(self, X, quantiles=0.5, method="nearest"):
"""Predict class or regression value for X at given quantiles.
Parameters
----------
X : {array-like, sparse matrix} of shape (n_samples, n_features)
Input data.
quantiles : float, optional
The quantiles at which to evaluate, by default 0.5 (median).
method : str, optional
The method to interpolate, by default 'linear'. Can be any keyword
argument accepted by :func:`~np.quantile`.
Returns
-------
y : ndarray of shape (n_samples, n_quantiles, [n_outputs])
The predicted values. The ``n_outputs`` dimension is present only
for multi-output regressors.
"""
return self.estimator_.predict_quantiles(X, quantiles, method)
[docs]
def get_leaf_node_samples(self, X):
"""Get samples in each leaf node across the forest.
Parameters
----------
X : array-like of shape (n_samples, n_features)
The data array.
Returns
-------
leaf_node_samples : array-like of shape (n_samples, n_estimators)
Samples within each leaf node.
"""
return self.estimator_.get_leaf_node_samples(X)
def _get_estimators_indices(self):
# Get drawn indices along both sample and feature axes
for tree in self.estimators_:
if not self.bootstrap and (
self._n_samples_bootstrap is None or (self._n_samples_bootstrap == self._n_samples)
):
yield np.arange(self._n_samples, dtype=np.int32)
else:
# tree.random_state is actually an immutable integer seed rather
# than a mutable RandomState instance, so it's safe to use it
# repeatedly when calling this property.
seed = tree.random_state
# Operations accessing random_state must be performed identically
# to those in `_parallel_build_trees()`
yield _generate_sample_indices(
seed, self._n_samples, self._n_samples_bootstrap, self.bootstrap
)
def _accumulate_prediction(predict, X, out, lock, indices=None):
"""
See https://github.com/scikit-learn/scikit-learn/blob/
95119c13af77c76e150b753485c662b7c52a41a2/sklearn/ensemble/_forest.py#L460
This is a utility function for joblib's Parallel.
It cannot be placed in ForestClassifier or ForestRegressor due to joblib's
compatibility issue with pickle.
"""
if indices is None:
indices = np.arange(X.shape[0])
proba = predict(X[indices], check_input=False)
with lock:
if len(out) == 1:
out[0][indices] += proba
else:
for i in range(len(out)):
out[i][indices] += proba[i]