将SHAP汇总图保存为PDF / SVG

Question

我目前正在处理分类问题，并希望创建特征重要性的可视化。我使用Python XGBoost包，它已经提供了功能重要性图。但是，我找到了shap（https://github.com/slundberg/shap），这是一个Python库，可以根据树分类器为特征重要性创建非常好的图。一切正常，我也可以将创建的图保存为PNG，但是，如果我尝试将其保存为PDF或SVG，我会得到一个例外。这是我在做的事情：

首先，我训练XGBoost模型，然后用bst表示模型。

train = remove_labels_for_binary_df(dataset_fc_baseline_1[0].train)
test = remove_labels_for_binary_df(dataset_fc_baseline_1[0].test)
results, bst = xgboost_with_bst(*transform_feat_to_num(train, test))

然后我创建shap值，使用这些值创建摘要图并保存创建可视化。如果我将情节保存为plt.savefig('shap.png')，一切正常。

import shap
import matplotlib.pyplot as plt

shap.initjs()

explainer = shap.TreeExplainer(bst)
shap_values = explainer.shap_values(train)
fig = shap.summary_plot(shap_values, train, show=False)
plt.savefig('shap.png')

但是，我需要PDF或SVG图而不是png，因此尝试用plt.savefig('shap.pdf')保存它，它通常工作正常，但会为shap图产生以下异常。

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-39-49d17973f438> in <module>()
  1 fig = shap.summary_plot(shap_values, train, show=False)
----> 2 plt.savefig('shap.pdf')

 C:\Users\Studio\Anaconda3\lib\site-packages\matplotlib\pyplot.py in 
savefig(*args, **kwargs)
708 def savefig(*args, **kwargs):
709     fig = gcf()
--> 710     res = fig.savefig(*args, **kwargs)
711     fig.canvas.draw_idle()   # need this if 'transparent=True' to reset 
colors
712     return res

C:\Users\Studio\Anaconda3\lib\site-packages\matplotlib\figure.py in 
savefig(self, fname, **kwargs)
2033             self.set_frameon(frameon)
2034 
-> 2035         self.canvas.print_figure(fname, **kwargs)
2036 
2037         if frameon:

C:\Users\Studio\Anaconda3\lib\site-packages\matplotlib\backend_bases.py in 
print_figure(self, filename, dpi, facecolor, edgecolor, orientation, format, 
**kwargs)
2261                 orientation=orientation,
2262                 bbox_inches_restore=_bbox_inches_restore,
-> 2263                 **kwargs)
2264         finally:
2265             if bbox_inches and restore_bbox:

C:\Users\Studio\Anaconda3\lib\site- 
packages\matplotlib\backends\backend_pdf.py in print_pdf(self, filename, 
**kwargs)
2584                 RendererPdf(file, image_dpi, height, width),
2585                 bbox_inches_restore=_bbox_inches_restore)
-> 2586             self.figure.draw(renderer)
2587             renderer.finalize()
2588             if not isinstance(filename, PdfPages):

C:\Users\Studio\Anaconda3\lib\site-packages\matplotlib\artist.py in 
draw_wrapper(artist, renderer, *args, **kwargs)
 53                 renderer.start_filter()
 54 
---> 55             return draw(artist, renderer, *args, **kwargs)
 56         finally:
 57             if artist.get_agg_filter() is not None:

C:\Users\Studio\Anaconda3\lib\site-packages\matplotlib\figure.py in 
draw(self, renderer)
1473 
1474             mimage._draw_list_compositing_images(
-> 1475                 renderer, self, artists, self.suppressComposite)
1476 
1477             renderer.close_group('figure')

C:\Users\Studio\Anaconda3\lib\site-packages\matplotlib\image.py in 
_draw_list_compositing_images(renderer, parent, artists, suppress_composite)
139     if not_composite or not has_images:
140         for a in artists:
--> 141             a.draw(renderer)
142     else:
143         # Composite any adjacent images together

C:\Users\Studio\Anaconda3\lib\site-packages\matplotlib\artist.py in 
draw_wrapper(artist, renderer, *args, **kwargs)
 53                 renderer.start_filter()
 54 
---> 55             return draw(artist, renderer, *args, **kwargs)
 56         finally:
 57             if artist.get_agg_filter() is not None:

C:\Users\Studio\Anaconda3\lib\site-packages\matplotlib\axes\_base.py in 
draw(self, renderer, inframe)
2605             renderer.stop_rasterizing()
2606 
-> 2607         mimage._draw_list_compositing_images(renderer, self, 
 artists)
2608 
2609         renderer.close_group('axes')

C:\Users\Studio\Anaconda3\lib\site-packages\matplotlib\image.py in 
_draw_list_compositing_images(renderer, parent, artists, suppress_composite)
139     if not_composite or not has_images:
140         for a in artists:
--> 141             a.draw(renderer)
142     else:
143         # Composite any adjacent images together

C:\Users\Studio\Anaconda3\lib\site-packages\matplotlib\artist.py in 
draw_wrapper(artist, renderer, *args, **kwargs)
 58                 renderer.stop_filter(artist.get_agg_filter())
 59             if artist.get_rasterized():
---> 60                 renderer.stop_rasterizing()
 61 
 62     draw_wrapper._supports_rasterization = True

C:\Users\Studio\Anaconda3\lib\site- 
packages\matplotlib\backends\backend_mixed.py in stop_rasterizing(self)
128 
129             height = self._height * self.dpi
--> 130             buffer, bounds = 
self._raster_renderer.tostring_rgba_minimized()
131             l, b, w, h = bounds
132             if w > 0 and h > 0:

C:\Users\Studio\Anaconda3\lib\site- 
packages\matplotlib\backends\backend_agg.py in tostring_rgba_minimized(self)
138                 [extents[0] + extents[2], self.height - extents[1]]]
139         region = self.copy_from_bbox(bbox)
--> 140         return np.array(region), extents
141 
142     def draw_path(self, gc, path, transform, rgbFace=None):

ValueError: negative dimensions are not allowed

你知道如何解决这个问题吗？提前致谢！

Answer 1

这是an issue between NumPy and matplotlib在使用rasterized=True（shap does if there are more than 500 datapoints）绘图时引起的，并且已在最新版本的matplotlib中得到解决。

Answer 2

以pdf格式保存：

plt.savefig("shap.pdf", format='pdf', dpi=1000, bbox_inches='tight')

保存为eps：

plt.savefig("shap.eps", format='eps', dpi=1000, bbox_inches='tight')

欲获得更多信息：

matplotlib.pyplot.savefig matplotlib

检查链接以了解更多信息，例如bbox_inches ='tight'的含义是什么。