Critical Description
DocumentCategorizerME no longer loads the tokenizer specified in the model manifest. Instead it always uses a WhitespaceTokenizer.
This appears to due to a change in 1.7.1 where the constructors for the DoccatFactory were modified to create a WhitespaceTokenizer.
This means the logic in the DoccatFactory.getTokenizer() method does try to load the tokenizer in model's manifest as "tokenizer" is not null when getTokenizer() is first called.