I was not referring to the things you mentioned. To ensure that OpenNLP works correctly all components have to be trained on larger data sets and then evaluated against a larger amount of test data. We often do changes where we expect the behavior of the component not to change.
In the past those tests where all manual and often took weeks to be finished, because they often showed signs of bugs which typically were hard to track down. Now we have a set of evaluation tests that can be run fully automatic. This means also that the tests are run more frequently during development, especially for high-risk changes, and we now should see problems before we merge new work into master. Jörn On Wed, Feb 1, 2017 at 2:59 PM, Richard Eckart de Castilho <[email protected]> wrote: > On 01.02.2017, at 14:35, Joern Kottmann <[email protected]> wrote: > > > > The project is now more agile and we can cut a release without a lot of > > overhead. We spent years working toward that goal. > > Now we will release a major version 1.x.0 every quarter and one or two > > minor 1.x.y versions every month. > > What did you do to minimize the release overhead? > > When voting on a release, I find myself double-checking license files, > changes in dependencies, checking signatures, etc. the most > time-consuming part (like minimum 1-2 hours). Did you somehow automatize > this (or maybe I'm just spending way too much time on details...). > > Cheers, > > -- Richard > >
