[Python] "Cannot convert a ndarray into a Tensor or Operation." error when trying to fetch a...

I have created a siamese network in tensorflow. I am calculating the distance between two tensors using the below code:

distance = tf.sqrt(tf.reduce_sum(tf.square(tf.subtract(question1_predictions, question2_predictions)), reduction_indices=1))


I am able to train the model without any errors. In the inference section, I am retrieving the distance tensor as below:

test_state, distance = sess.run([question1_final_state, distance], feed_dict=feed)


Tensorflow is throwing an error:

Fetch argument array([....], dtype=float32) has invalid type , must be a string or Tensor. (Can not convert a ndarray into a Tensor or Operation.)
​

When I print the distance tensor, before and after the session.run in the training section, it shows as <class 'tensorflow.python.framework.ops.Tensor'>. So the replacement of tensor distance with numpy distance is happening in the session.run of inference section. Following the code from the inference section:

with graph.as_default():
saver = tf.train.Saver()

with tf.Session(graph=graph) as sess:
sess.run(tf.global_variables_initializer(), feed_dict={embedding_placeholder: embedding_matrix})
saver.restore(sess, tf.train.latest_checkpoint('checkpoints'))

test_state = sess.run(initial_state)

for ii, (x1, x2, batch_test_ids) in enumerate(get_test_batches(test_question1_features, test_question2_features, test_ids, batch_size), 1):
feed = {question1_inputs: x1,
question2_inputs: x2,
keep_prob: 1,
initial_state: test_state
}
test_state, distance = sess.run([question1_final_state, distance], feed_dict=feed)

Continue reading...