Let's say we are using mini-batch gradient descent in tensorflow. Everyday I update last day's model to predict something(regression/classification) by feeding new data to the model. What do I do if a new value for a categorical feature shows up? How do I incorporate it in the existing model?
For example, let's say a feature called state had 3 values till yesterday namely "CA", "IA" and "VA". So the input feature vector in my tensorflow model had three dummy variables- each for one of the three states. Now when I reuse this model today using a new set of data, how do I accomodate for a new value for the feature state, say, for example "NC". How do I change the feature set in the existing model?