Skip to content
Browse files

Better document how to return multiple values from forward (#457)

* Better document how to return multiple values from forward

Add an elaborate example to the Avanced_Usage.ipynb
notebook. Furthermore, add a short entry in the FAQ that also links to
the notebook.

* Address reviewer comments.

* Address reviewer comments
  • Loading branch information...
BenjaminBossan authored and ottonemo committed Jun 11, 2019
1 parent 177a2a8 commit 815ade33a8b2374ef0f3206dfc4b5cbd1ee1e2ba
Showing with 507 additions and 98 deletions.
  1. +23 −1 docs/user/FAQ.rst
  2. +484 −97 notebooks/Advanced_Usage.ipynb
@@ -60,7 +60,7 @@ that, try to wrap your dictionary into a :class:`.SliceDict`. This is
a data container that partly behaves like a dict, partly like an
ndarray. For more details on how to do this, have a look at the
coresponding `data section
in the notebook.

@@ -246,3 +246,25 @@ The ``RandomDataset`` can be passed directly to
net = NeuralNet(MyModule, criterion=torch.nn.MSELoss)
How can I deal with multiple return values from forward?

skorch supports modules that return multiple values. To do this,
simply return a tuple of all values that you want to return from the
``forward`` method. However, this tuple will also be passed to the
criterion. If the criterion cannot deal with multiple values, this
will result in an error.

To remedy this, you need to either implement your own criterion that
can deal with the output or you need to override
:func:`` and handle the unpacking of the

To inspect all output values, you can use either the
:func:`` method (eager) or the
:func:`` method (lazy).

For an example of how this works, have a look at this `notebook

0 comments on commit 815ade3

Please sign in to comment.
You can’t perform that action at this time.