Help with Art_Generation_with_Neural_Style_Transfer UNQ_C5

Course Q&A Deep Learning Specialization Convolutional Neural Networks
1

I’m really unsure of where I went wrong. I have restarted the kernal and ran from the beginning, and checked other postings. It appears that I’m using a tensor that cannot be converted to a numpy array.

NotImplementedError: in user code:

File "<ipython-input-127-4f32495065ab>", line 19, in train_step  *
    J_style = compute_style_cost(a_S, a_G)
File "<ipython-input-53-7b56545257a6>", line 30, in compute_style_cost  *
    J_style_layer = compute_layer_style_cost(a_S[i], a_G[i])
File "<ipython-input-47-b68791c415c8>", line 27, in compute_layer_style_cost  *
    J_style_layer = 1/(4*n_C**2*n_H**2*n_W**2)*tf.reduce_sum(np.square(GS-GG))

NotImplementedError: Cannot convert a symbolic tf.Tensor (sub:0) to a numpy array. This error may indicate that you're trying to pass a Tensor to a NumPy call, which is not supported.
2

he error you’re seeing suggests that you’re trying to pass a TensorFlow tensor object to a NumPy function that only works with NumPy arrays. Can you try to replace the np.square function in compute_layer_style_cost with the tf.square function, which computes the element-wise square of a tensor:

    J_style_layer = 1/(4*n_C**2*n_H**2*n_W**2)*tf.reduce_sum(tf.square(GS-GG))
3

Well, you can convert TF tensors to numpy arrays, but the point is that you need to explicitly specify that, as in:

myTensor.numpy()

But it doesn’t just happen implicitly. So you could have written that with np.square using the “explicit conversion” technique, but you’d have to convert it back to a tensor after the np.square. But what is the point when @JazzKaur has pointed out a much cleaner solution. :nerd_face:

4

Oh my gosh. Thank you! I looked back through my code 1000x, and I guess I used np.square out of habit, and just glossed over it every time thinking it was tf.square. Thank you!

5

This is tricky because the only way some of us got out of the compute_layer_style_cost was by changing tf.square to np.square. Going back to that function and changing it to tf square has an exception that comes up regarding: x and y must have the same dtype, got tf.float32 != tf.int32.
It’s like a sinking boat, you plug one hole and then a different one pops up.

6

It is deep water. We have to be careful with everything. Even 2. and 2 are not the same. Single dot can create a massive change.

BTW, if you are facing any error, feel free to share your error with us.

Best,
Saif.