I want to create a utility function where I give it a tensor t
and a variable name n
, and the function returns (it it exists) the variable
whos name contains n
and is part of the graph of t
.
def get_variable(t, n):
#code
return variable
The reason I want to do this is that using Jupyter I often create a graph multiple times while deciding the best structure, but the tensor names keep changing to something like: {name}_{repetition}:0
, so finding them through tf.all_variables()
becomes increasingly harder.
It would be easier if I could limit the search to the variables related to a specific tensor because its identifier is of the latest repetition.
Aucun commentaire:
Enregistrer un commentaire