Categorical preprocessing layers are a new feature in Keras (available in tf-nightly & TF 2.3). They enable you to do string/int feature encoding directly as part of your model -- one-hot, multi-hot, indexing, hashing, etc. Check out this simple example:https://keras.io/examples/structured_data/structured_data_classification_from_scratch/ …
-
It’s very explicit, which i like, but quite verbose compared to TF FeatureColumn. I think FeatureColumn is super unintuitive especially when you see it for the first time, and I wonder if there is a middle ground of concise structured input representation and readability?
Preprocessing layers are fairly low-level since they expose every step of the preprocessing. You could always write a higher-level, more FeatureColumn-like interface on top. E.g. the `encoder_*` functions in this example are probably close to that: https://keras.io/examples/structured_data/structured_data_classification_from_scratch/#feature-preprocessing-with-keras-layers …
-
-
I know, I’m wondering if there’s room for this in the public API.
-
Also, something like input_layer and sequence_input_layer, which take in a features dict and feature columns, can abstract away a lot of unnecessary complexity.
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.