How to solve the data dependency problem when building the computing graph?

[eljrte]

When I am building a computing graph and creating two tensors eg. A & B . tensor B's shape(ne) is dependent on A's data member, which is unknown before compute_forward the graph and is illegal to reach when building graph.
How can I deal with such cases? As I want to be more dynamic in building the graph. Thanks a lot.

[JohannesGaessler]

I think it is not intended for tensor shapes to be set dynamically during the compute graph evaluation. A branching evaluation based on some condition would maybe be possible but this is not implemented (and may be seen as adding too much complexity). What you can do is create two separate compute graphs and change how you construct the second one based on the results of the first one.

[eljrte]

I think it is not intended for tensor shapes to be set dynamically during the compute graph evaluation. A branching evaluation based on some condition would maybe be possible but this is not implemented (and may be seen as adding too much complexity). What you can do is create two separate compute graphs and change how you construct the second one based on the results of the first one.

Thanks a lot. And is it possible to build nested compute graphs? Like I have a "big" compute graph, and I build and run a smaller compute graph inside it, using its result in the big graph?

[JohannesGaessler]

No, and I think this is not intended either.