You can leverage built-in functionality like the progress bar, callbacks, performance optimizations (like step fusing)... while training a GAN. In a few lines. At no extra effort.pic.twitter.com/krrArIQGZH
Show this thread
A Keras workflow that's new in TF 2.2 is the ability to leverage `fit()` with a custom learning algorithm, by overriding `train_step`. (Of course, you can do the same with `test_step` too.)pic.twitter.com/SDWRLPKmDA
-
-
-
Show this thread
End of conversation
New conversation -
-
-
My experience with tf tutorials is that they work fine (at least for 6 months, until the next API change). But when you intend to do something slightly different, all hell breaks loose.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
wow this is awesome
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
This is incredibly helpful, thank you!
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
That's exactly the feature that I missed from tf keras! Operating through a high level api with minimum effort while being able to customize the training in detail if necessary, love it!
-
This Tweet is unavailable.
- Show replies
New conversation -
-
-
I don’t fully understand the use of GradientTape inside train_step. Is this implementation figuring out how to calculate gradients at each training step? If so, wouldn’t it be more efficient to find the computation graph for the gradients once at the beginning, and reuse it?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
This is fantastic. Thanks for all the great work Francois. You have literally made my job a thousand times easier.
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
That is AWSOME!
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
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.