How you know you nailed an API design
Left: how you would plot the Xception architecture in a paper. Right: how you would implement it with the Functional API (that's the entire code). 1:1 match between how you think about it and how you write it.pic.twitter.com/ehCo1jirGv
-
-
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
This Tweet is unavailable.
-
If I am right, the whole python notebook thing is also inspired by Mathematica notebook?
- Show replies
-
-
-
I wish it was easier to draw this diagrams integrated into the ide or jupyter notebook
-
If you call `plot_model` on the `xception` model instance in this notebook, you get this: https://s3-us-west-1.amazonaws.com/fchollet-public-colabs/xception.png … It's too large for use in a paper, but it's great for debugging complex architectures.
End of conversation
New conversation -
-
-
In tensorlang we went one step further and eliminated the need for `x = ...(x)`. So you only need to name the values that are used in multiple places. Have you thought about doing something like that in keras?
-
I've simulated the tensorlang syntax in Python by overriding __rshift__ or __or__. Checking if the second arg is callable helps avoid the need to manually "call" each function directly. Much less boilerplate that way IMO.
End of conversation
New conversation -
-
-
wow very interesting
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
are these three functions referentially transparent?
-
As far as the user is concerned, yes. Side effects are purely internal. Only user-visible side-effect is incrementing the counter used for automatic layer name generation (but these names are purely cosmetic, and this goes away if you use explicit names for all your layers)
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.