When you're using tf.keras in TensorFlow 2.0, training your models with `fit` is one option, but you can also write a custom training loop. It takes about 10 lines.pic.twitter.com/Tn3NA4pZaf
Show this thread
If your model creates additional losses during its forward pass (i.e. if you called `self.add_loss` in a custom layer), collect them in one line like this.pic.twitter.com/tuvBdibMAP
-
-
The advantage of using `fit` is that it packs a lot of functionality in a convenient interface: a progress bar, easy reporting of metrics such as precision & recall, easy management of class weights for imbalanced classification, callbacks for model checkpointing and more...pic.twitter.com/pPwETGqZvw
Show this threadThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Is there a suggested way to incorporate callbacks in these custom training loops? I could only find this suggested way, which looks appropriate. https://groups.google.com/forum/#!topic/keras-users/IS-wBskAg0k …
-
That's correct, you can always call the callback methods manually at different stages of training (e.g. `on_epoch_begin`, `on_epoch_end`...)
- Show replies
New conversation -
-
-
I'm admittedly a beginner here, but wouldn't you want the tape var to be used inside the with block? It seems like this way it's not being disposed of, unless that's desired?
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Thanks for the guide, it would be awesome if you could provide us with the Google Colab.
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.