Closed emlautarom1 closed 3 years ago
Classifiers could be generated at startup of a duckling instance and then taken as an input (they're just hashmaps after all), but this isn't done for one reason or another. I don't know what that reason is. For now, having classifiers checked in is just easier and leads to faster startup times, but as you say, at the cost of more expensive diffs. For me, I usually just minimise the classifier files since they're 99.999999% noise when reviewing. Our internal review tool typically recognises them as large/mostly noise and minimises them automatically.
I was wondering why are classifiers kept under version control, given that they are auto generated using
duckling-regen-exe
. Is there an actual reason or is this just a coincidence?. I ask mainly since this introduces some noise when committing and reviewing PRs (I'm currently learning about the codebase)