The current space/time tradeoffs are great! Keep those if at all possible. A few extra seconds of startup time is not a big deal, but a very long thesaurus initialization is not really acceptable.
One option would be to use the shelving database I created elsewhere, and simply generate it on the first run.
The existing thesaurus system sucks.
Create a new one based on this: https://raw.githubusercontent.com/drichert/moby/develop/share/moby/thesaurus/roget13a.txt
The current space/time tradeoffs are great! Keep those if at all possible. A few extra seconds of startup time is not a big deal, but a very long thesaurus initialization is not really acceptable.
One option would be to use the shelving database I created elsewhere, and simply generate it on the first run.