Post by 2020 in Graphomata comments

Viewing post in Graphomata comments

20203 years ago

The UI issue in the charts should be fixed by now - extreme values of scoring thresholds will affect scaling only to a limited extent. After this change, do you still feel that log scale is important for conveying the information?

By some parts of the dotted lines, do you mean that tooltips don't activate when you hover anywhere over the bar or the line chart? Or did you try hovering the space between them (which indeed doesn't currently have a tooltip)? These tooltips should list all accepted thresholds (were added a month ago, together with a 'show: max' option).

Regarding blacklisted thresholds, I've just added them back in order to validate the UI, but the roadmap is to treat brute-force solutions as rooted - for scoring and in the histograms. It will take a while to get there, so for now I guess they're here (despite their effect on the scores of nearby nonoptimal solutions).

L0laapk33 years ago (3 edits)

(Edited: added section on the bottom)
The new scale seems to help for the most part except when you have 'extreme' solutions of your own :P Thats not really a problem I guess, while it does still mess with the other solutions it is also quite satisfying to see an absolutely huge cost :P Overall I think its fine currently, and while I personally still am a fan of log plots I don't think it is necessary at this point.

About the treshold list tooltip: I was referring to the part inbetween the two graphs. Overall I think they deserve to be promoted to a permanent spot on the stats page, considering their importance.

For marking the 'extreme' solutions as rooted. I don't think it will be that simple, since it's not really a black and white thing. It can span the whole range from unrolling one iteration with 1 extra command all the way to some pretty extreme cases (I have seen dom's E7 no-longer-optimal intermediate scores that scale pretty well all the way until where he is now.

Edit: it would also be nice to be able to see the achieved tradeoffs in root (I'm not 100% sure that makes any sense in the current context). Currently you can kind of guess what happened based on the other graphs, but theres no guarantee. Particularly on G4 I am struggling, I even ended up finding a new best non-root solution that made my rooted solution take quite a hit on the upgrade but it feels like the highest root solution wasn't affected as much as mine, it would be nice to see its tradeoff as to know in what direction to search :

Edit edit: the dotted lines turn green when you've achieved them. It would be a nice touch to make them gold when they are an unique best since there is currently no way to tell which are unique :)

20203 years ago

The plan is to detect non-general solutions that assume some limit on input sizes. This will be done by adding larger inputs as I used to do before, except that they'll be hidden and failure on them will just make the solution rooted.

Since players can't select those extra inputs, I can use only a few and make them exceedingly large without slowing down rendering or testing. Note that loop unrolling, as long as it supports arbitary input sizes, will not make a solution rooted.

Regarding your suggestions: uniqueness information can't be overlaid on the existing charts as is, because the client and the server may have different thresholds. And as you implied, best-known rooted solutions don't affect scoring so they don't contribute thresholds. Technically everything can be presented (maybe in a new screen), but it's going to be tricky to explain the meaning of this information to players.

itch.io

Viewing post in Graphomata comments