Minimum loss reduction required to make a further partition on a leaf node of the tree. The larger gamma is, the more conservative the algorithm will be.

  • Usually "conservative" algorithms are ones that more tightly match the expected results. They have less fuzziness, so they return fewer results. That may or may not be a good thing, as you can filter out otherwise good matches. – Berin Loritsch Mar 4 at 13:58
  • 4
    Where is this used? What algorithm is your example referring to, or what domain-space does it pertain to? – mmathis Mar 4 at 14:17
  • 5
    Honestly, this question is hard to answer in a sensible manner without more context. "conservative" can mean a dozen of very different things in different contexts. – Doc Brown Mar 4 at 14:55

The term conservative is rather ambiguous. I would not recommend to use it to describe the behavior of an algorithm, since it is not clear what conservation it is about.

From the context, it appears that your algorithm could be about balancing trees or compression. It also appears that the gamma factor influences how much more conservative it is as opposed to reduction or losses. So this seems to be about changes in the data structure that the algorithm manages. More conservative means less changes (e.g. less rebalancing of nodes or less loss due to compression).

Another frequent meaning is about novelty. You could say that an algorithm is more conservative when it implements well-known methods, is composed of proven building blocks, and is less innovative.

But again, looking at the context, here it’s certainly about the preservation of the state of the data structure, so that it remains as close as possible to its original state.

Not the answer you're looking for? Browse other questions tagged or ask your own question.