Structured data preprocessing layers include: - CategoryEncoding: do one-hot/multi-hot/count/tf-idf integer categorical feature encoding - StringLookup/IntegerLookup: build indexes of categorical features - Hashing: the hashing trick - Discretization: continuous feature binning
-
-
Show this thread
-
- Normalization: featurewise normalization - CategoryCrossing: generate co-occurence features of categorical features Together, these layers enable arbitrarily complex preprocessing pipelines that make your models truly end-to-end, from raw data to predictions.
Show this thread -
This helps reduce the training/serving skew: since all the preprocessing is part of the model, you can load and use your model without having to be aware of how each feature is expected to be encoded & normalized.
Show this thread
End of conversation
New conversation -
-
-
This is fascinating! Traditionally, you would apply a weighting to the different variables/ parameters because not all elements are equal. I wonder what is the delta in prediction accuracy....

-
This is a neural network: the importance of each feature is learned by the model during training.
End of conversation
New conversation -
-
-
This looks like the UCI heart disease dataset
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
This is great. I really wish, though, that we had a centralized library for preprocessing that worked for TF, Keras, Scikit, PyTorch, big/medium/small data, everything...oh to dream...
-
In general, you can use these layers and extract the Numpy outputs to use with another framework, e.g. layer = CategoryEncoding() layer.adapt(np_data) # Learn the index preprocessed_np_data = layer(np_data).numpy() # Convert the input
End of conversation
New conversation -
-
-
Thx a lot. Please add works only with tf-nightly
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Bookmarked for the weekend. This sounds so cool. Also that model plot is

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.