[FIXED] Cannot convert a symbolic Keras input/output to a numpy array. when using model.optimizer.get_gradients in Tensorflow 2.4

Issue

I followed the tutorial A Primer on Deep Learning in Genomics - Public.ipynb in colab but got TypeError: Cannot convert a symbolic Keras input/output to a numpy array... as I tried to execute the step 4.Interpret at line sal = compute_salient_bases(model, input_features[sequence_index]).

import tensorflow.keras.backend as K

def compute_salient_bases(model, x):
  input_tensors = [model.input]
  gradients = model.optimizer.get_gradients(model.output[0][1], model.input)
  compute_gradients = K.function(inputs = input_tensors, outputs = gradients)
  
  x_value = np.expand_dims(x, axis=0)
  gradients = compute_gradients([x_value])[0][0]
  sal = np.clip(np.sum(np.multiply(gradients,x), axis=1),a_min=0, a_max=None)
  return sal

sequence_index = 1999  # You can change this to compute the gradient for a different example. But if so, change the coloring below as well.
sal = compute_salient_bases(model, input_features[sequence_index])

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-10-b6400cc2276d> in <module>()
      1 sequence_index = 1999  # You can change this to compute the gradient for a different example. But if so, change the coloring below as well.
----> 2 sal = compute_salient_bases(model, input_features[sequence_index])
      3 
      4 plt.figure(figsize=[16,5])
      5 barlist = plt.bar(np.arange(len(sal)), sal)

14 frames
/usr/local/lib/python3.7/dist-packages/tensorflow/python/keras/engine/keras_tensor.py in __array__(self)
    272   def __array__(self):
    273     raise TypeError(
--> 274         'Cannot convert a symbolic Keras input/output to a numpy array. '
    275         'This error may indicate that you\'re trying to pass a symbolic value '
    276         'to a NumPy call, which is not supported. Or, '

TypeError: Cannot convert a symbolic Keras input/output to a numpy array. This error may indicate that you're trying to pass a symbolic value to a NumPy call, which is not supported. Or, you may be trying to pass Keras symbolic inputs/outputs to a TF API that does not register dispatching, preventing Keras from automatically converting the API call to a lambda layer in the Functional Model.

The problem is at model.optimizer.get_gradients(model.output[0][1], model.input). I think it is the right usage according to https://www.tensorflow.org/api_docs/python/tf/gradients

get_gradients(
    loss, params
)

I am quite confused about the error. Or is there an alternative approach to compute_salient_bases?

Solution

Downgrade TensorFlow version, restart runtime and run the notebook again.

!pip install tensorflow==1.13.2
import tensorflow as tf
print(tf.__version__)

Answered By - fuad021