Is is possible to have a `tf.keras.layer` with an internal non trainable `tf.Variable` with `None` shape that is updated during training? @TensorFlo @TensorFlow @fchollet
Variables cannot have dynamic shapes, since they contain actual values. For every new shape you want, you have to create a new variable. It's possible for a layer to leverage different variables at each batch if your layer is dynamic. Pass a `dynamic=True` layer constructor arg.
-
-
Please, check our website and mobile app, as we badly need
#AI#MachineLearning#DataScience community support! Website - https://ainews.page IOS - https://apps.apple.com/us/app/ainews-page-ai-ml-ar-news/id1501119075 … Android - https://play.google.com/store/apps/details?id=page.ainews.app …Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
thanks! so in the `build` I should create all the variables for the different possible shapes and then in the call use the right one according to the inputs shape? Indeed I'd like to maintain sort of an internal state based on the received batches
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.