Gradient export

[ZofiaTr]

Hi Frederik,

I am testing the class for gradient exporting from TATi. When I run my python code in the terminal, everything works fine, but when I do the same in the jupyter notebook I got this error:

Create model for initial optimization
[<tf.Tensor 'walker1/gradients/walker1/layer1/Wx_plus_b/MatMul_grad/MatMul_1:0' shape=(2, 2) dtype=float32>]
[<tf.Tensor 'walker1/gradients_1/walker1/layer1/Wx_plus_b/add_grad/Reshape_1:0' shape=(2,) dtype=float32>]
[<tf.Tensor 'walker1/gradients_2/walker1/layer2/Wx_plus_b/MatMul_grad/MatMul_1:0' shape=(2, 1) dtype=float32>]
[<tf.Tensor 'walker1/gradients_3/walker1/layer2/Wx_plus_b/add_grad/Reshape_1:0' shape=(1,) dtype=float32>]
[<tf.Tensor 'walker1/gradients_4/walker1/output/Wx_plus_b/MatMul_grad/MatMul_1:0' shape=(1, 1) dtype=float32>]
[<tf.Tensor 'walker1/gradients_5/walker1/output/Wx_plus_b/add_grad/Reshape_1:0' shape=(1,) dtype=float32>]

.....

FailedPreconditionError (see above for traceback): GetNext() failed because the iterator has not been initialized. Ensure that you have run the initializer operation for this iterator before getting the next element.
[[Node: IteratorGetNext = IteratorGetNextoutput_shapes=[[?,2], [?,1]], output_types=[DT_FLOAT, DT_INT32], _device="/job:localhost/replica:0/task:0/device:CPU:0"]]

Do you have any idea what is going on? It it not critical, I can always run it in the terminal..

[FrederikHeber]

As a first guess I would say that you do not reset the dataset iterator, i.e. this needs to happen before calling train().

import tati.models.model as model
...
model.reset_dataset()

Does this help?

[ZofiaTr]

No. It is the same. I can show it from terminal..