One more thing. You can do automated shape inference. Less cognitive load for the user.pic.twitter.com/8LYV81XCAt
Show this thread
This is the Chainer API. It's nice, but I have two suggestions. 1) no need for `init_scope`, you can instead override `__setattr__` to automatically register layers. 2) the activation function should be a layer argument. Chaining plain layer calls looks nicer. @beam2dpic.twitter.com/hcyrD9vvrc
-
New conversation
-
-
-
Is in keras or what?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
I disagree with 2): I think it makes the flow of the network easier to understand when I only have to read the forward call and not jump back to the init to see which (or even if an) activation is used.
-
Probably Linear(x).relu() looks the nicest
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.