Plot imodels trees with dtreeviz.
dtreeviz draws a ShadowDecTree; imodels models are built from scikit-learn
trees underneath, so one can be built with dtreeviz's own ShadowSKDTree
(https://github.com/parrt/dtreeviz). dtreeviz is not a dependency of imodels and
is imported only when this is called.
Expand source code
"""Plot imodels trees with dtreeviz.
dtreeviz draws a `ShadowDecTree`; imodels models are built from scikit-learn
trees underneath, so one can be built with dtreeviz's own `ShadowSKDTree`
(https://github.com/parrt/dtreeviz). dtreeviz is not a dependency of imodels and
is imported only when this is called.
"""
import numpy as np
from imodels.util.model_trees import n_tree_outputs, sklearn_trees
def shadow_tree(model, X, y, feature_names=None, target_name='target',
class_names=None, tree_num=0):
"""Build a dtreeviz ShadowSKDTree for one of a model's trees.
Parameters
----------
model
A fitted tree-based imodels model (FIGS, hierarchical shrinkage, CART,
TAO, boosted rules, ...).
X, y : the data to annotate the tree with, as dtreeviz requires.
feature_names : list of str, optional
Defaults to the names the model was fitted with, else X0, X1, ... .
target_name : str
Name shown for the target.
class_names : list, optional
Classification only. Defaults to the model's ``classes_``.
tree_num : int
Which tree to draw, for models made of several (FIGS, boosted rules).
Returns
-------
dtreeviz.models.sklearn_decision_trees.ShadowSKDTree
Raises
------
ImportError
If dtreeviz is not installed.
ValueError
If the model is not tree-based, or tree_num is out of range.
Examples
--------
>>> import dtreeviz # doctest: +SKIP
>>> from imodels import FIGSClassifier, shadow_tree # doctest: +SKIP
>>> model = FIGSClassifier(max_rules=5).fit(X, y) # doctest: +SKIP
>>> viz = dtreeviz.trees.DTreeVizAPI( # doctest: +SKIP
... shadow_tree(model, X, y, tree_num=0))
>>> viz.view() # doctest: +SKIP
"""
try:
from dtreeviz.models.sklearn_decision_trees import ShadowSKDTree
except ImportError as error: # pragma: no cover - depends on the environment
raise ImportError(
"dtreeviz is needed to plot imodels trees but is not installed "
"('pip install dtreeviz'). It is not a dependency of imodels."
) from error
trees = sklearn_trees(model)
if trees is None:
raise ValueError(
f"Don't know how to draw {type(model).__name__}. dtreeviz support "
"covers tree-based models; if the model is not fitted yet, fit it first."
)
if not 0 <= tree_num < len(trees):
raise ValueError(
f"tree_num must be in [0, {len(trees)}) for this model, got {tree_num}."
)
X = np.asarray(X)
if feature_names is None:
feature_names = _feature_names(model, X.shape[1])
if class_names is None:
class_names = _class_names(model)
return ShadowSKDTree(trees[tree_num], X, np.asarray(y),
list(feature_names), target_name, class_names)
def _feature_names(model, n_features):
for attr in ('feature_names_in_', 'feature_names_', 'feature_names'):
names = getattr(model, attr, None)
if names is not None and len(names) == n_features:
return [str(name) for name in names]
return [f'X{i}' for i in range(n_features)]
def _class_names(model):
"""dtreeviz wants class labels for classifiers and None for regressors."""
classes = getattr(model, 'classes_', None)
if classes is None or n_tree_outputs(model) == 1:
return None
return [str(label) for label in classes]Functions
def shadow_tree(model, X, y, feature_names=None, target_name='target', class_names=None, tree_num=0)-
Build a dtreeviz ShadowSKDTree for one of a model's trees.
Parameters
model- A fitted tree-based imodels model (FIGS, hierarchical shrinkage, CART, TAO, boosted rules, …).
- X, y : the data to annotate the tree with, as dtreeviz requires.
feature_names:listofstr, optional- Defaults to the names the model was fitted with, else X0, X1, … .
target_name:str- Name shown for the target.
class_names:list, optional- Classification only. Defaults to the model's
classes_. tree_num:int- Which tree to draw, for models made of several (FIGS, boosted rules).
Returns
dtreeviz.models.sklearn_decision_trees.ShadowSKDTree
Raises
ImportError- If dtreeviz is not installed.
ValueError- If the model is not tree-based, or tree_num is out of range.
Examples
>>> import dtreeviz # doctest: +SKIP >>> from imodels import FIGSClassifier, shadow_tree # doctest: +SKIP >>> model = FIGSClassifier(max_rules=5).fit(X, y) # doctest: +SKIP >>> viz = dtreeviz.trees.DTreeVizAPI( # doctest: +SKIP ... shadow_tree(model, X, y, tree_num=0)) >>> viz.view() # doctest: +SKIPExpand source code
def shadow_tree(model, X, y, feature_names=None, target_name='target', class_names=None, tree_num=0): """Build a dtreeviz ShadowSKDTree for one of a model's trees. Parameters ---------- model A fitted tree-based imodels model (FIGS, hierarchical shrinkage, CART, TAO, boosted rules, ...). X, y : the data to annotate the tree with, as dtreeviz requires. feature_names : list of str, optional Defaults to the names the model was fitted with, else X0, X1, ... . target_name : str Name shown for the target. class_names : list, optional Classification only. Defaults to the model's ``classes_``. tree_num : int Which tree to draw, for models made of several (FIGS, boosted rules). Returns ------- dtreeviz.models.sklearn_decision_trees.ShadowSKDTree Raises ------ ImportError If dtreeviz is not installed. ValueError If the model is not tree-based, or tree_num is out of range. Examples -------- >>> import dtreeviz # doctest: +SKIP >>> from imodels import FIGSClassifier, shadow_tree # doctest: +SKIP >>> model = FIGSClassifier(max_rules=5).fit(X, y) # doctest: +SKIP >>> viz = dtreeviz.trees.DTreeVizAPI( # doctest: +SKIP ... shadow_tree(model, X, y, tree_num=0)) >>> viz.view() # doctest: +SKIP """ try: from dtreeviz.models.sklearn_decision_trees import ShadowSKDTree except ImportError as error: # pragma: no cover - depends on the environment raise ImportError( "dtreeviz is needed to plot imodels trees but is not installed " "('pip install dtreeviz'). It is not a dependency of imodels." ) from error trees = sklearn_trees(model) if trees is None: raise ValueError( f"Don't know how to draw {type(model).__name__}. dtreeviz support " "covers tree-based models; if the model is not fitted yet, fit it first." ) if not 0 <= tree_num < len(trees): raise ValueError( f"tree_num must be in [0, {len(trees)}) for this model, got {tree_num}." ) X = np.asarray(X) if feature_names is None: feature_names = _feature_names(model, X.shape[1]) if class_names is None: class_names = _class_names(model) return ShadowSKDTree(trees[tree_num], X, np.asarray(y), list(feature_names), target_name, class_names)