Shap: plot_cmap not working in all cases

[gabrielcs]

I am trying to customize the force_plot colors with Seaborn but I am getting the following exceptions, according to whether I use as_cmap=True:

plot_cmap=sns.diverging_palette(15, 155, s=99, l=70, n=2)

results in:

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
shap/plots/force.py in force_plot(base_value, shap_values, features, feature_names, out_names, link, plot_cmap)
     92             DenseData(np.zeros((1, len(feature_names))), list(feature_names))
     93         )
---> 94         return visualize(e, plot_cmap)
     95 
     96     else:

shap/plots/force.py in visualize(e, plot_cmap)
    188 
    189 def visualize(e, plot_cmap="RdBu"):
--> 190     plot_cmap = verify_valid_cmap(plot_cmap)
    191     if isinstance(e, AdditiveExplanation):
    192         return AdditiveForceVisualizer(e, plot_cmap=plot_cmap).html()

shap/plots/force.py in verify_valid_cmap(cmap)
    183         _rgbstring = re.compile(r'#[a-fA-F0-9]{6}$')
    184         for color in cmap:
--> 185              assert(bool(_rgbstring.match(color))),"Invalid color found in CMAP."
    186 
    187     return cmap

TypeError: cannot use a string pattern on a bytes-like object

And setting as_cmap=True

plot_cmap=sns.diverging_palette(15, 155, s=99, l=70, n=2, as_cmap=True)

results in:

---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
shap/plots/force.py in force_plot(base_value, shap_values, features, feature_names, out_names, link, plot_cmap)
     92             DenseData(np.zeros((1, len(feature_names))), list(feature_names))
     93         )
---> 94         return visualize(e, plot_cmap)
     95 
     96     else:

shap/plots/force.py in visualize(e, plot_cmap)
    188 
    189 def visualize(e, plot_cmap="RdBu"):
--> 190     plot_cmap = verify_valid_cmap(plot_cmap)
    191     if isinstance(e, AdditiveExplanation):
    192         return AdditiveForceVisualizer(e, plot_cmap=plot_cmap).html()

shap/plots/force.py in verify_valid_cmap(cmap)
    178 def verify_valid_cmap(cmap):
    179     assert (isinstance(cmap, str) or isinstance(cmap, list) or str(type(cmap)).endswith("unicode'>")
--> 180         ),"Plot color map must be string or list! not: " + str(type(cmap))
    181     if isinstance(cmap, list):
    182         assert (len(cmap) > 1), "Color map must be at least two colors."

AssertionError: Plot color map must be string or list! not: <class 'matplotlib.colors.LinearSegmentedColormap'>

[slundberg]

Ah. The problem here is that force_plot uses javascript and D3 for rendering, not matplotlib, so you need to pass a list of colors to use. Pass a list of two colors to replace the default red and blue as hex strings (or some other string valid in CSS). You could probably get these colors from seaborn by evaluating their cmap at the values you want and passing the result as a list.

[ghost]

Hi, this still appears to not be working:
display(shap.force_plot(expected_val, shap_vals[shap_index_list[0],:], X.iloc[shap_index_list[0],:], matplotlib=True,text_rotation=30, link='logit', show = True, plot_cmap=['#00bfff','#ffffff']))

I receive no error, however the default colors remain.

[lakraj]

'plot_cmap' for force_plots doesn't work for me as well. I need to convert the colors to a grey and black combination.

I tested with the examples given in - https://shap.readthedocs.io/en/latest/example_notebooks/tree_explainer/Force%20Plot%20Colors.html.
It plots the defaults colors.