Tensorflow Object to a numpy array

Question

I am trying to use a custom F1 score function in Keras but my target variable is an array and I don't want to one hot encode them, so I am using sparse_categorical_crossentropy as a loss.

from keras import backend as K
def f1(y_true, y_pred):
  print(type(y_true))
  return y_true

When I run the above code I am getting y_true and y_pred as <class 'tensorflow.python.framework.ops.Tensor'>. How can we convert y_true and y_pred to NumPy array, so that I can implement sklearn's F1 score function upon them?

I am getting this following error. ```AttributeError: 'Tensor' object has no attribute 'numpy'``` — Maagalam HARSHA VARDHAN, Jun 28 '20 at 13:48
I'm not a TF user, but I think only eager tensors have this attribute. — hpaulj, Jun 28 '20 at 16:25
There are lots of SO questions with this error in the title, e.g. https://stackoverflow.com/questions/52357542/attributeerror-tensor-object-has-no-attribute-numpy — hpaulj, Jun 28 '20 at 16:31

score 0 · Answer 1 · answered Jun 28 '20 at 12:48

0

Use (as also suggested by @learner in comments):

y_true = y_true.numpy()
y_pred = y_pred.numpy()

answered Jun 28 '20 at 12:48

Ehsan

12,072
2
20
33

Tensorflow Object to a numpy array

1 Answers1