It would be nice in some cases if there was the option to save out feature vectors from sequence files when processed using a particular feature specification. A possible scenario is when doing cross fold validation through the CLI interface it would not need to process each raw text sequence into the respective feature set. Questions over how these could be aggregated later on remain for example would they all need to be generated in one pass from all candidate source documents so as to have a consistent feature index into each vector etc. Also any numerical issues regarding any feature normalisation that may occur.
It would be nice in some cases if there was the option to save out feature vectors from sequence files when processed using a particular feature specification. A possible scenario is when doing cross fold validation through the CLI interface it would not need to process each raw text sequence into the respective feature set. Questions over how these could be aggregated later on remain for example would they all need to be generated in one pass from all candidate source documents so as to have a consistent feature index into each vector etc. Also any numerical issues regarding any feature normalisation that may occur.