DOCS: Notebooks adjoint about firedrake HOT 6 CLOSED

In general the adjoint shouldn't play well with notebooks due to the possibilities for out-of-order execution. However for this case would it be sufficient to reset the tape in one of the uppermost cells? That would also be a convenient place to warn the user about the perils of running the cells in the wrong order.

from firedrake.

I tried

     get_working_tape().clear_tape()
     get_working_tape().reset_blocks()

That was not sufficient to fix this failure.

I agree with you about the warning.

from firedrake.

I think I found a solution. If you replace continue_annotating() at the top of the notebook with

if not annotate_tape():
    continue_annotation()

then it seems to work. I think this is because of the way annotation is tracked with an integer rather than a boolean so calling continue_annotation() twice actually breaks things.

from firedrake.

I think I found a solution. If you replace continue_annotating() at the top of the notebook with

if not annotate_tape():
    continue_annotation()

then it seems to work. I think this is because of the way annotation is tracked with an integer rather than a boolean so calling continue_annotation() twice actually breaks things.

That sounds like a bug, did you report it? I can't see anything here or here

from firedrake.

I think I found a solution. If you replace continue_annotating() at the top of the notebook with

if not annotate_tape():
    continue_annotation()

then it seems to work. I think this is because of the way annotation is tracked with an integer rather than a boolean so calling continue_annotation() twice actually breaks things.

That sounds like a bug, did you report it? I can't see anything here or here

See dolfin-adjoint/pyadjoint#111

from firedrake.

This issue is sorted with pyadjoint PR 113

from firedrake.