Playing around with Tensorflow Transform for making input pipelines
A few notes:
- Can only use TFT in Python 2.7! Per this and this, waiting on Python 3 support for
apache-beam - Per this, you may need to run
sudo apt-get install libsnappy-devbefore installingpython-snappy
===
While I have found TFT super-useful, I am still constrained by preprocessing that can be done with native TF ops since TFT exports a TF Graph. I haven't found ways to do common text preprocessing steps, e.g. lowercase, Porter stemming, removing stop words, with TF/TFT.