I don’t know how to fix this problem “No gradients provided for any variable”
I tried "model.trainable_weights " and “model.trainable_variables”. But i still get “No gradients provided for any variable” error.
{mentor edit: image of code removed}
1 Like
Please click my name and message your notebook as an attachment.
1 Like
I’m having what might be the exact same issue. Was there a resolution to this?
Printed output from my C2W3_Assignment notebook debugging:
loss : tf.Tensor(0.6931472, shape=(), dtype=float32)
trainable_variables : <class ‘list’>
trainable_variables[0]: (2, 64)
trainable_variables[1]: (64,)
trainable_variables[2]: (64, 1)
trainable_variables[3]: (1,)
about to run grads = tape.gradient(loss, model.trainable_variables)
grads: [None, None, None, None]
Error from optimizer is :
ValueError: No gradients provided for any variable: [‘dense_15/kernel:0’, ‘dense_15/bias:0’, ‘dense_16/kernel:0’, ‘dense_16/bias:0’].
1 Like
Are you still involved with this? Any chance I could get you to take a look?
Thanks for asking. I’m not the right person to take up advanced tensorflow assignment test cases at this point.
1 Like
Hello,
I dont know exactly where the problem might be either but send me your code on PM , I will have a look on it later on.
1 Like
Right, I saw you assignment and I see that when you set the losses (in the “set_sparse_cat_crossentropy_loss”) you are not doing it properly, you need to use the losses class not metrics class!
Try after changing that and lets see what happens.
2 Likes
That was it… thanks very much. I failed to notice the difference between:
<class ‘tensorflow.python.keras.metrics.SparseCategoricalCrossentropy’>
vs.
<class ‘tensorflow.python.keras.losses.SparseCategoricalCrossentropy’>
T.
2 Likes
Thank you so much Sir
2 Likes