fix: UnboundLocalError with bar plots using dictionary input

CHANGELOG.md

@@ -22,6 +22,8 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
   ([dsgibbons#88](https://github.com/dsgibbons/shap/pull/88) by @thatlittleboy).
 - Fixed sampling in `shap.datasets` to sample without replacement
   ([dsgibbons#36](https://github.com/dsgibbons/shap/pull/36) by @thatlittleboy).
+- Fixed an `UnboundLocalError` problem arising from passing a dictionary input to `shap.plots.bar`
+  ([#3001](https://github.com/slundberg/shap/pull/3000) by @thatlittleboy).
 - Fixed tensorflow import issue with Pyspark when using `Gradient`
   ([#2983](https://github.com/slundberg/shap/pull/2983) by @skamdar).
 - Fixed deprecation warnings for `numpy>=1.24` from numpy types

shap/plots/_bar.py

@@ -10,6 +10,7 @@
 
 from .. import Cohorts, Explanation
 from ..utils import format_value, ordinal_str
+from ..utils._exceptions import DimensionError
 from . import colors
 from ._labels import labels
 from ._utils import (
@@ -54,17 +55,35 @@ def bar(shap_values, max_display=10, order=Explanation.abs, clustering=None, clu
         cohorts = {"": shap_values}
     elif isinstance(shap_values, Cohorts):
         cohorts = shap_values.cohorts
+    elif isinstance(shap_values, dict):
+        cohorts = shap_values
     else:
-        assert isinstance(shap_values, dict), "You must pass an Explanation object, Cohorts object, or dictionary to bar plot!"
+        emsg = (
+            "The shap_values argument must be an Explanation object, Cohorts "
+            "object, or dictionary of Explanation objects!"
+        )
+        raise TypeError(emsg)
 
     # unpack our list of Explanation objects we need to plot
     cohort_labels = list(cohorts.keys())
     cohort_exps = list(cohorts.values())
-    for i in range(len(cohort_exps)):
-        if len(cohort_exps[i].shape) == 2:
-            cohort_exps[i] = cohort_exps[i].abs.mean(0)
-        assert isinstance(cohort_exps[i], Explanation), "The shap_values paramemter must be a Explanation object, Cohorts object, or dictionary of Explanation objects!"
-        assert cohort_exps[i].shape == cohort_exps[0].shape, "When passing several Explanation objects they must all have the same shape!"
+    for i, exp in enumerate(cohort_exps):
+        if not isinstance(exp, Explanation):
+            emsg = (
+                "The shap_values argument must be an Explanation object, Cohorts "
+                "object, or dictionary of Explanation objects!"
+            )
+            raise TypeError(emsg)
+
+        if len(exp.shape) == 2:
+            # collapse the Explanation arrays to be of shape (#features,)
+            cohort_exps[i] = exp.abs.mean(0)
+        if cohort_exps[i].shape != cohort_exps[0].shape:
+            emsg = (
+                "When passing several Explanation objects, they must all have "
+                "the same number of feature columns!"
+            )
+            raise DimensionError(emsg)
         # TODO: check other attributes for equality? like feature names perhaps? probably clustering as well.
 
     # unpack the Explanation object

tests/plots/test_bar.py

@@ -1,17 +1,82 @@
-''' This file contains tests for the bar plot.
-'''
+"""This file contains tests for the bar plot.
+"""
 import matplotlib.pyplot as plt
+import numpy as np
 import pytest
 
 import shap
+from shap.utils._exceptions import DimensionError
+
+
+@pytest.mark.parametrize(
+    "unsupported_inputs",
+    [
+        [1, 2, 3],
+        (1, 2, 3),
+        np.array([1, 2, 3]),
+        {"a": 1, "b": 2},
+    ],
+)
+def test_input_shap_values_type(unsupported_inputs):
+    """Check that a TypeError is raised when shap_values is not a valid input type."""
+    emsg = (
+        "The shap_values argument must be an Explanation object, Cohorts "
+        "object, or dictionary of Explanation objects!"
+    )
+    with pytest.raises(TypeError, match=emsg):
+        shap.plots.bar(unsupported_inputs, show=False)
+
+
+def test_input_shap_values_type_2():
+    """Check that a DimensionError is raised if the cohort Explanation objects have different shape."""
+    rs = np.random.RandomState(42)
+    emsg = (
+        "When passing several Explanation objects, they must all have "
+        "the same number of feature columns!"
+    )
+    with pytest.raises(DimensionError, match=emsg):
+        shap.plots.bar(
+            {
+                "t1": shap.Explanation(
+                    values=rs.randn(40, 10),
+                    base_values=np.ones(40) * 0.5,
+                ),
+                "t2": shap.Explanation(
+                    values=rs.randn(20, 5),
+                    base_values=np.ones(20) * 0.5,
+                ),
+            },
+            show=False,
+        )
 
 
 @pytest.mark.mpl_image_compare
-def test_simple_bar(explainer): # pylint: disable=redefined-outer-name
-    """ Check that the bar plot is unchanged.
-    """
+def test_simple_bar(explainer):
+    """Check that the bar plot is unchanged."""
     shap_values = explainer(explainer.data)
     fig = plt.figure()
     shap.plots.bar(shap_values, show=False)
     plt.tight_layout()
     return fig
+
+
+@pytest.mark.mpl_image_compare
+def test_simple_bar_with_cohorts_dict():
+    """Ensure that bar plots supports dictionary of Explanations as input."""
+    rs = np.random.RandomState(42)
+    fig = plt.figure()
+    shap.plots.bar(
+        {
+            "t1": shap.Explanation(
+                values=rs.randn(40, 5),
+                base_values=np.ones(40) * 0.5,
+            ),
+            "t2": shap.Explanation(
+                values=rs.randn(20, 5),
+                base_values=np.ones(20) * 0.5,
+            ),
+        },
+        show=False,
+    )
+    plt.tight_layout()
+    return fig