def assert_tree_classifier_explained(clf, vec, target_names, **explain_kwargs):
top = 30
get_res = lambda: explain_weights(
clf, vec=vec, target_names=target_names, top=top, **explain_kwargs)
res = get_res()
expl_text, expl_html = format_as_all(res, clf)
for expl in [expl_text, expl_html]:
assert 'feature importances' in expl
assert 'god' in expl # high-ranked feature
if len(res.feature_importances.importances) > top:
assert 'more features' in expl or 'more …' in expl
if isinstance(clf, (DecisionTreeClassifier, OneVsRestClassifier)):
if _graphviz.is_supported():
assert '<svg' in expl_html
else:
assert '<svg' not in expl_html
assert res == get_res()
def test_explain_tree_classifier(newsgroups_train, clf):
docs, y, target_names = newsgroups_train
vec = CountVectorizer()
X = vec.fit_transform(docs)
clf.fit(X.toarray(), y)
top = 30
get_res = lambda: explain_weights(
clf, vec=vec, target_names=target_names, top=top)
res = get_res()
expl_text, expl_html = format_as_all(res, clf)
for expl in [expl_text, expl_html]:
assert 'feature importances' in expl
assert 'god' in expl # high-ranked feature
if len(res.feature_importances.importances) > top:
assert 'more features' in expl or 'more …' in expl
if isinstance(clf, (DecisionTreeClassifier, OneVsRestClassifier)):
if _graphviz.is_supported():
assert '<svg' in expl_html
else:
assert '<svg' not in expl_html
assert res == get_res()
def _format_decision_tree(treedict):
if treedict.graphviz and _graphviz.is_supported():
return _graphviz.dot2svg(treedict.graphviz)
else:
return tree2text(treedict)
