So I am trying to implement SHAP to a CNN model I trained on an image dataset on Kaggle. But I keep getting the following Attribute Error:

---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[9], line 1
----> 1 explainer = shap.DeepExplainer(s_model, val_data)
      2 sp_values = explainer.shap_values(val_data)
      4 shap.summary_plot(sp_values, val_data)

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:155, in TFDeep.__init__(self, model, data, session, learning_phase_flags)
    153     self.expected_value = None
    154 else:
--> 155     if self.data[0].shape[0] > 5000:
    156         warnings.warn("You have provided over 5k background samples! For better performance consider using smaller random sample.")
    157     if not tf.executing_eagerly():

AttributeError: '_PrefetchDataset' object has no attribute 'shape'

I am using the following code to split the dataset:

from tensorflow.keras.utils import image_dataset_from_directory

batch_size = 32
img_height = 224
img_width = 224

train_data = image_dataset_from_directory(
                  data_dir,
                  validation_split=0.3,
                  subset="training",
                  seed=123,
                  image_size=(img_height, img_width),
                  batch_size=batch_size)


val_data = image_dataset_from_directory(data_dir,
                                        validation_split=0.3,
                                        subset="validation",
                                        seed=123,
                                        image_size=(img_height,img_width),
                                        batch_size=batch_size) 

After loading a previously trained model, I use the following code:

explainer = shap.KernelExplainer(s_model, val_data)
sp_values = explainer.shap_values(val_data)

shap.summary_plot(sp_values, val_data)

This is where I get the error.

I am using Tensorflow and Keras for the code. Can anyone help me solve this? I want to implement SHAP, LIME and any other interpretability method on some image models.