Shap: Saving SHAP plots programmatically in Python

[MaxPowerWasTaken]

First off, thanks a lot for such an awesome tool!

I think I might be missing something obvious, but I'm trying to save SHAP plots from Python, that I'm displaying with the shap plotting functions. I tried a couple ways:

import matplotlib.pyplot as plt
...
shap.summary_plot(shap_values, final_model_features)
plt.savefig('scratch.png')

and...

import matplotlib.pyplot as plt
...
fig = shap.summary_plot(shap_values, final_model_features)
plt.savefig('scratch.png')

but each just saves a blank image. Is there something obvious I'm missing to programmatically save these awesome plots from Python? Or should I just be re-generating them in matplotlib off the SHAP values matrix to do that? Thanks!

[slundberg]

By default summary_plot calls plt.show() to ensure the plot displays.
But if you pass show=False to summary_plot then it won't, and think
that might fix the blank figure issue you are having.

On Thu, Jul 12, 2018 at 7:36 AM Max Epstein [e-mail]
wrote:

First off, thanks a lot for such an awesome tool!

I think I might be missing something obvious, but I'm trying to save SHAP
plots from Python, that I'm displaying with the shap plotting functions. I
tried a couple ways:

import matplotlib.pyplot as plt
...
shap.summary_plot(shap_values, final_model_features)
plt.savefig('scratch.png')

and...

import matplotlib.pyplot as plt
...
fig = shap.summary_plot(shap_values, final_model_features)
plt.savefig('scratch.png')

but each just saves a blank image. Is there something obvious I'm missing
to programmatically save these awesome plots from Python? Or should I just
be re-generating them in matplotlib off the SHAP values matrix to do that?
Thanks!

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
https://github.com/slundberg/shap/issues/153, or mute the thread
https://github.com/notifications/unsubscribe-auth/ADkTxceFgvnQkAAYwNiea4Tb4u3aLlF2ks5uF17jgaJpZM4VM-FL
.

[MaxPowerWasTaken]

hey thanks! that did it.

[JoshKoh]

shap.summary_plot(shap_values, X, show=False)
This is not working for me. The plot still shows and plt.savefig only gives blank.

[slundberg]

@JoshKoh perhaps you have something like %matplotlib inline in your notebook? That might change things.

[hmanz]

Is there a way to store the figure into a variable instead of saving it to file?
Thanks

[slundberg]

@hmanz after running shap.summary_plot(shap_values, X, show=False) you can run import matplotlib.pyplot as pl; f = pl.gcf() to get the current figure in the variable f. What you do with it after that depends on matplotlib and not shap.

[hmanz]

@slundberg thank you for your great solution! I was able to use that solution for all SHAP plots except for the multiple-sample force-plots since there is no support for matplotlib=True. Is there a way to accomplish the same task for the multiple-sample force-plots?

More specifically, I am trying to store the plot as a variable so it can be handled by another class's method that will save it to file as a png or jpeg.

[slundberg]

Multiple-sample force-plots are Javascript/HTML. So in order to save those programmatically you will need to have some kind of browser-like renderer. Those exist but I can't comment on all the details to get them running.

[wabc1994]

Why that not work for me?

My code just like the follow, @slundberg
the code just come from your official demo ,and i just want to save the demo image

import xgboost
import shap
import matplotlib.pyplot as plt
# load JS visualization code to notebook
shap.initjs()
# train XGBoost model
X,y = shap.datasets.boston()
model = xgboost.train({"learning_rate": 0.01}, xgboost.DMatrix(X, label=y), 100)

# explain the model's predictions using SHAP values
# (same syntax works for LightGBM, CatBoost, and scikit-learn models)
explainer = shap.TreeExplainer(model)
shap_values = explainer.shap_values(X)

# visualize the first prediction's explanation (use matplotlib=True to avoid Javascript)
# fig = plt.gcf()
# shap._*plot本质上就是调用 plt.show()， 所有我们的思路就是如何在下面的文件进行保存
shap.force_plot(explainer.expected_value, shap_values[0,:], X.iloc[0,:],show=False)
plt.savefig('scratch.png')

[dzieciou]

@wabc1994 I was able to save your plot by replacing the last two lines with:

shap.force_plot(explainer.expected_value, shap_values[0,:], 
    X.iloc[0,:],show=False,matplotlib=True)
    .savefig('scratch.png')

Basically, it enables rendering with matplotlib instead of JavaScript and calls savefig on figure returned by force_plot().

[paris007]

,show=False,matplotlib=True)
.savefig('scratch.png')

shap.force_plot(explainer.expected_value, shap_values, X,show=False,matplotlib=True).savefig('scratch.png')

I tried the code above and I get this error:
Exception: matplotlib = True is not yet supported for force plots with multiple samples!

[sagydr]

plt.savefig worked for me. Thanks!
but.. can't I save it to HTML like I can after running force_plot?

shap.save_html('explainer.html', shap.force_plot(...))

[slundberg]

Note that when matplotlib=True is set then you can't save an HTML version because no HTML version exists (it is now a matplotlib figure, which is typically best exported as a PDF not HTML)

[charlatteD]

shap.force_plot(explainer.expected_value, shap_values[0,:],
X.iloc[0,:],show=False,matplotlib=True)
.savefig('scratch.png')

This works for me. But by specifying "matplotlib" = True, the resolution of the plot was downgraded, and a more serious issue is some parts of the original plot were cropped. Anyone had a similar issue?

[bradblanch]

@charlatteD this should solve your resolution and cropping issue:
shap.force_plot(explainer.expected_value, shap_values[0,:], X.iloc[0,:],show=False,matplotlib=True).savefig('scratch.png',format = "png",dpi = 150,bbox_inches = 'tight')

[Frederiklux]

I can save my force plots with:
shap.force_plot(explainer.expected_value[1], shap_values[1][0,:], top_11.iloc[0,:],show=False,matplotlib=True).savefig('test.png')

But when I include link="logit" as an argument, I get "posx and posy should be finite values" a bunch of times and a blank plot. Does anyone know how to solve this?

[lhang33]

I can save my force plots with:
shap.force_plot(explainer.expected_value[1], shap_values[1][0,:], top_11.iloc[0,:],show=False,matplotlib=True).savefig('test.png')

But when I include link="logit" as an argument, I get "posx and posy should be finite values" a bunch of times and a blank plot. Does anyone know how to solve this?

Same with you. Only that I found it occurs in version 0.36.0, but works well in 0.35.0. Hope it will be fixed sooner.

[giangnguyen2412]

When using plot_image, I can not save image because it does not return anything. I think plotting in shap is quite poor! Does someone have any experiences?

[TrinhDinhPhuc]

I got this error for force_plot Exception: matplotlib = True is not yet supported for force plots with multiple samples!

[Z-Shakeri]

The following workaround did the trick for me:

I first exported the interactive chart to HTML, then I saved the HTML file to a high-resolution pdf file.

Here is the code snippet I used:

f=shap.force_plot(explainer.expected_value, shap_values, X, show=False)
shap.save_html("index.htm", f)