Currently the LogRegIntentClassifier uses all the CPU on the machine it's training while the ProbabilisticIntentClassifier trains it's CRFSlotFillers sequentially.
Expected behavior
Ideally:
the CRFSlotFiller training should be parallelized
the number of jobs for the intent classification and the slot filling should be configurable
Current behavior
Currently the
LogRegIntentClassifieruses all the CPU on the machine it's training while theProbabilisticIntentClassifiertrains it'sCRFSlotFillers sequentially.Expected behavior
Ideally:
CRFSlotFillertraining should be parallelized