TFDV: Schema for LSTM


I am trying to apply my learned knowledge about TFX to an external problem with the goal of developing an LSTM to classify strings. I have ingested my dataset from tfrecord files and am now creating a schema (cutout):

 	                  Type 	Presence 	Valency 	Domain
Feature name 			
'name' 	             STRING 	required 		'name_domain'

with an empty name_domain domain.

When analyzing for anomalies I get the anomaly “Unexpected string values”.

How can I create a schema with TFDV where I can describe arbitrary strings as a feature for my LSTM?

Please go back to course 2 week 1 ungraded lab to see how anomalies are addressed in string fields. New values are added to race column in the lab.