shap/shap

BUG: LookupError: gradient registry has no entry for: shap_TensorListStack

Opened this issue · 3 comments

AmishFaldu commented

Issue Description

When finding shap values for the deep learning model using Tensorflow, there is some operation lookup error for following operations:

  1. TensorListStack
  2. BatchMatMulV2 - Traceback is similar to TensorListStack
  3. While - Traceback is similar to TensorListStack
  4. TensorListFromTensor - Traceback is similar to TensorListStack

Can you please help me on this?

Thank you

Minimal Reproducible Example

import tensorflow as tf
import shap

inputs = Input(shape=(series_len, n_features,))
lstm_out = GRU(10, return_sequences=True, dropout=0.5, recurrent_dropout=0.3)(inputs)
hidden_states_t = Permute((2, 1))(lstm_out)
hidden_size = int(lstm_out.shape[2])
hidden_states_t = Reshape((hidden_size, series_len))(hidden_states_t)
score_first_part = Dense(series_len, use_bias=False)(hidden_states_t)
score_first_part_t = Permute((2, 1))(score_first_part)
lambda_out = Lambda(lambda x: x[:, :, -1], output_shape=(hidden_size, 1))(hidden_states_t)
score = dot([score_first_part_t, lambda_out], [2, 1])
output = Dense(1, activation='sigmoid', activity_regularizer=tf.keras.regularizers.l2(0.0001))(score)

model = tf.keras.Model(inputs=inputs, outputs=output)
model.build((32, series_len, n_features))
model.summary()

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

Traceback

---------------------------------------------------------------------------
StagingError                              Traceback (most recent call last)
Cell In[11], line 13
      8 explainer = shap.DeepExplainer(
      9     model,
     10     X_train
     11 )
     12 # shap_values = explainer.shap_values(X_test)
---> 13 shap_values = explainer.shap_values(X_test[:10])

File ~/.venv/lib/python3.11/site-packages/shap/explainers/_deep/__init__.py:135, in DeepExplainer.shap_values(self, X, ranked_outputs, output_rank_order, check_additivity)
     91 def shap_values(self, X, ranked_outputs=None, output_rank_order='max', check_additivity=True):
     92     """Return approximate SHAP values for the model applied to the data given by X.
     93 
     94     Parameters
   (...)
    133 
    134     """
--> 135     return self.explainer.shap_values(X, ranked_outputs, output_rank_order, check_additivity=check_additivity)

File ~/.venv/lib/python3.11/site-packages/shap/explainers/_deep/deep_tf.py:310, in TFDeep.shap_values(self, X, ranked_outputs, output_rank_order, check_additivity)
    308 # run attribution computation graph
    309 feature_ind = model_output_ranks[j,i]
--> 310 sample_phis = self.run(self.phi_symbolic(feature_ind), self.model_inputs, joint_input)
    312 # assign the attributions to the right part of the output arrays
    313 for t in range(len(X)):

File ~/.venv/lib/python3.11/site-packages/shap/explainers/_deep/deep_tf.py:366, in TFDeep.run(self, out, model_inputs, X)
    363         tf_execute.record_gradient = tf_backprop.record_gradient
    365     return final_out
--> 366 return self.execute_with_overridden_gradients(anon)

File ~/.venv/lib/python3.11/site-packages/shap/explainers/_deep/deep_tf.py:401, in TFDeep.execute_with_overridden_gradients(self, f)
    399 # define the computation graph for the attribution values using a custom gradient-like computation
    400 try:
--> 401     out = f()
    402 finally:
    403     # reinstate the backpropagatable check
    404     if hasattr(tf_gradients_impl, "_IsBackpropagatable"):

File ~/.venv/lib/python3.11/site-packages/shap/explainers/_deep/deep_tf.py:359, in TFDeep.run.<locals>.anon()
    357     v = tf.constant(data, dtype=self.model_inputs[i].dtype)
    358     inputs.append(v)
--> 359 final_out = out(inputs)
    360 try:
    361     tf_execute.record_gradient = tf_backprop._record_gradient

File ~/.venv/lib/python3.11/site-packages/tensorflow/python/util/traceback_utils.py:153, in filter_traceback.<locals>.error_handler(*args, **kwargs)
    151 except Exception as e:
    152   filtered_tb = _process_traceback_frames(e.__traceback__)
--> 153   raise e.with_traceback(filtered_tb) from None
    154 finally:
    155   del filtered_tb

File ~/.venv/lib/python3.11/site-packages/tensorflow/python/eager/polymorphic_function/autograph_util.py:52, in py_func_from_autograph.<locals>.autograph_handler(*args, **kwargs)
     50 except Exception as e:  # pylint:disable=broad-except
     51   if hasattr(e, "ag_error_metadata"):
---> 52     raise e.ag_error_metadata.to_exception(e)
     53   else:
     54     raise

StagingError: in user code:

    File "~/.venv/lib/python3.11/site-packages/shap/explainers/_deep/deep_tf.py", line 249, in grad_graph  *
        x_grad = tape.gradient(out, shap_rAnD)

    LookupError: gradient registry has no entry for: shap_TensorListStack

Expected Behavior

It should return the shap values

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

0.45.0

MZZheng commented

same issue.

shap: 0.46.0
tf: 2.17.0