BUG: 'tuple' object has no attribute 'as_list'

[Amirhossein4096]

Issue Description

I used shap to interpret my cnn model. When shap tries to get the output shape in its internal function, i get error. I know it's because of the tensorflow version, how can I fix this error?
There isn't problem in tensorflow==2.15.0
Please solve this problem for tensorflow==2.16.1

Minimal Reproducible Example

import shap
explainer = shap.DeepExplainer(model=model, data = X_train[0:10])
shap_values = explainer.shap_values(X_test[0:10])

Traceback

AttributeError
Traceback (most recent call last)
<ipython-input-9-5d9f53dcf3ce> in <cell line: 1>()
----> 1 explainer = shap.DeepExplainer(model=model, data = X_train[0:10])
      2 shap_values = explainer.shap_values(X_test[0:10])
1 frames
/usr/local/lib/python3.10/dist-packages/shap/explainers/_deep/deep_tf.
py in __init__(self, model, data, 
session, learning_phase_flags)
    170             self.phi_symbolics = [None]
    171         else:
--> 172             noutputs = self.model_output.shape.as_list()[1]
    173             if noutputs is not None:
    174                 self.phi_symbolics = [None for i in range(noutputs)]

AttributeError: 'tuple' object has no attribute 'as_list'

Expected Behavior

No response

Bug report checklist

• I have checked that this issue has not already been reported.
• I have confirmed this bug exists on the latest release of shap.
• I have confirmed this bug exists on the master branch of shap.
• I'd be interested in making a PR to fix this bug

Installed Versions

my package versions:
tensorflow==2.16.1
shap==0.45.1

[CloseChoice]

Thanks for reporting @Amirhossein4096 . Would you be so kind and provide an example that reproduces the issue? Your code is not complete since we cannot define the model in it. Would be great if you could trim your code down to the minimal steps needed to produce this bug.

[NanK77]

I am using same versions of TensorFlow and Shap. With a LSTM model, I've received same error:

AttributeError Traceback (most recent call last)
Cell In[51], line 11
9 background=shap.sample(scaled_train_X,100)
10 # Create the SHAP DeepExplainer using the model and the reshaped background dataset
---> 11 explainer = shap.DeepExplainer(model, background)
13 # Calculate SHAP values for the reshaped test set
14 shap_values = explainer.shap_values(scaled_test_X)

File ~\AppData\Local\anaconda3\Lib\site-packages\shap\explainers_deep_init_.py:90, in DeepExplainer.init(self, model, data, session, learning_phase_flags)
87 super().init(model, masker)
89 if framework == 'tensorflow':
---> 90 self.explainer = TFDeep(model, data, session, learning_phase_flags)
91 elif framework == 'pytorch':
92 self.explainer = PyTorchDeep(model, data)

File ~\AppData\Local\anaconda3\Lib\site-packages\shap\explainers_deep\deep_tf.py:172, in TFDeep.init(self, model, data, session, learning_phase_flags)
170 self.phi_symbolics = [None]
171 else:
--> 172 noutputs = self.model_output.shape.as_list()[1]
173 if noutputs is not None:
174 self.phi_symbolics = [None for i in range(noutputs)]

AttributeError: 'tuple' object has no attribute 'as_list'

[CloseChoice]

@NanK77 would you be so kind to share code so that we can reproduce the issue? We need a definition of the data (at best a minimal set of row) and code to construct the model.

[macayla-donegan]

I'm also running into this issue- I can't really share code that reproduces the issue because I work in health tech, but the inputs to shap.DeepExplainer are a 3 layer model trained in tensorflow:
simple_nn = Sequential()
simple_nn.add(Dense(256, activation='relu', input_shape=(len(self.feature_names),)))
simple_nn.add(Dense(256, activation='relu'))
simple_nn.add(Dropout(0.2))
simple_nn.add(Dense(256, activation='relu'))
simple_nn.add(Dense(1, activation='sigmoid'))

and a dataframe of numerical features with no nans

can someone let me know if they found a workaround for this?

Here's the exact error I'm getting

AttributeError Traceback (most recent call last)
Cell In[60], line 3
1 import shap
----> 3 explainer = shap.DeepExplainer(nn_model, X_train)
4 shap_values = explainer.shap_values(X_test)

File /opt/conda/lib/python3.10/site-packages/shap/explainers/_deep/init.py:84, in DeepExplainer.init(self, model, data, session, learning_phase_flags)
81 framework = 'tensorflow'
83 if framework == 'tensorflow':
---> 84 self.explainer = TFDeep(model, data, session, learning_phase_flags)
85 elif framework == 'pytorch':
86 self.explainer = PyTorchDeep(model, data)

File /opt/conda/lib/python3.10/site-packages/shap/explainers/_deep/deep_tf.py:172, in TFDeep.init(self, model, data, session, learning_phase_flags)
170 self.phi_symbolics = [None]
171 else:
--> 172 noutputs = self.model_output.shape.as_list()[1]
173 if noutputs is not None:
174 self.phi_symbolics = [None for i in range(noutputs)]

AttributeError: 'tuple' object has no attribute 'as_list'

[CloseChoice]

Sorry, should have known this before, but we currently do not support tensorflow 2.16 right now. See https://github.com/shap/shap/blob/master/pyproject.toml#L74. Will close this here, since support of tf 2.16 and keras 3 is way more complicated than this here. See the mentioned issue #3672