NitroML is a modular, portable, and scalable model-quality benchmarking framework for Machine Learning and Automated Machine Learning (AutoML) pipelines.
Apache License 2.0
41
stars
6
forks
source link
Custom transform component for flexibility in `preprocessing_fn` #14
Implemented a custom component that extends from the default tfx Tranform allowing a custom signature for preprocessing_fn .
We use the new transform component to automate processing of OpenML datasets. In the current implementation, we compute the vocab of label column which is used in the Trainer component for label transformation (string -> integer) as required for the Keras model.
preprocessing_fn
.