So SnowiestDrowners is similar but not 100% anymore of course, giving it a try while waiting to see further seed analysis from @JoeDaFrogman but I have no clue if the POI placement is the same and it is too soon to tell (guess I should have done a creative flythrough). I'd love to know more of your methodology since it seems like you've got a much better understanding of it than most.
@
Red Eagle LXIX
Ok, steel yourselves, this is going to be a very long reply.
The first question you have to ask is "what makes a good seed?" If someone posted 2,3,10 seeds in a single comment how would you "KNOW" objectively which is the best seed and which is the worst? The answer is you just can't unless your answer is "this one looks pretty" and that's ok as every data point is always subjective. My thought process is to remove at least SOME of the subjectivity and inject some objectivity, ie things you can put a concrete measurement to.
Lets look at A19 seeds for example to start with. Say I were to somehow(pretend with me here) have the EXACT same image file for the same seed name/size with the same cities in the same locations everything was exactly the same except one difference. Both have largish cites centralized near 0,0 and coincidentally this is roughly near where you chose to put your living base. From the picture you could not tell one from the other.
Now if I add in the FACT that for picture 1, 4 of the tier 5's were in one of those two cities(and thus very close by to you) but for the picture 2 NONE of the tier 5's were in those two cities and you had to travel 3.5 k to any of the tier 5's AND those 4 tier 5's were all in opposite corners of the map(and thus in different biomes from each other). So out of those two seeds, which would be preferable? I suspect that for the vast majority of people, unless they REALLY enjoy the exploration element of the game(and there as some who do!), that picture 1 would be the seed the would prefer to have.
Now, being that there is always some subjectivity, the goal is to try to reduce that as much as possible(I will admit, this this moves the subjectively back into the hands of the program author!). I am working the
@dcsobral, who has written and refined over a few alpha's a set of tools to generate worlds in a mostly automated way. He has evolved this quite a bit until A20. At this point, he has a working rating for each seed based on (and this is a super high level overview as I do not know all of the variables):
For a given list of POI's(traders, tier 3, tier 4, tier 5's, etc) locate and rank each POI(currently each POI has the same ranking, possibly changing in a future version!) and find the one spot on the entire map in radius(currently 1km from a central point) with the highest rating of unique POIs with the rating criteria(ie, regular houses with nothing in them count for nothing.) We are currently generating random worlds automatically using this technique and throwing away any that don't meet an arbitrary cut off, while saving the remaining into a zip file of the world plus some extras. We are finding that given a roughly 12-13 hour run of 100 seeds(8k) will produce about 12 or so that are above that cut off. This then gives us something to look over both objectively by reviewing the number(and which ones) of tier 4's or tier 5's, etc are within that 1km from the best base location and then a subjective assessment of "does the map look pretty, where is the best base location, etc". For me, I won't hold out of hand for someone else a best base location in the wasteland, but "I" won't use that seed.
How does this all work? It uses the windows subsystem for linux as currently many of the "programs" are shell scripts. Dan(and I to some extend) are working on converting these to use Python with an ultimate goal to possibly make a Python based GUI that can be run from a windows machine. Note that there are some other third party programs required as well AND most importantly, you cant be using that computer while you are running the scripts. In any event, once the scripts are completed the world generation, there are some other scripts that can then help you analyze some of the data that is saved for each world and rank them according to each other.
Now, in A19, we did NOT have a rating, but I did Perl script to take all the seeds and analyze and it spit out something like the below. I then simply sorted by the top15 POI's(this is an arbitrary list Dan and a number of others determined were the most desirable POI's) based on unique count. Note that this was the results of somewhere near 3000 seeds generated that took a bit over a month of my computer doing nothing for long stretch's at a time:
This program no longer works in A20 due to the file format changes the world gen put into place, so I am taking this opportunity to port the logic to python(which I am learning) as well as to add in the rating and a few other goodies such as the best base location coordinates, etc. My first attempt at this was just the information from column J and to the left and with Dan's help we eventually added the biome percentages since for some people that's important information to know.
I am extremely happy that Dan now has the rating system, as again it removes even MORE subjectivity and mainly as that means I don't have to write it (lol). One of the thoughts he had recently was to possibly add a "forest rating" as well as the overall rating, again where the rating is "find the one point with the highest rating within radius of X". If he does add, this would then bring back some subjectively to the system as how would you sort those on a spreadsheet with two data points? Our goal is not to say "here is THE best seed" but to help narrow down to a curated list of some good seeds so someone does have to wade through COMPLETELY subjective "here is a good seed". We both belong to a YTer's community so a big part of our goal is to help support that community with some "good" seeds.
For reference, I have a person I play with on occasion and toward the end of A19, she said "we are starting a new world", She told me she got the seed from a video some random person made and when she told me the seed, I looked it up and found it was on I had created! I watched the video and two of the 15 on the list were ones I had created, so I was quite proud/happy/smug(not really sure what I should have felt).
If you are VERY technical, especially with some linux knowledge, then his code is located here:
https://github.com/dcsobral/seedGen. Ill be honest, set up is a PITA. And the less you know about linux the worse it would be and not advisable to use. the processing tools can create a high res image with the idea base location pin pointed on a copy of the map, as well a thumbnail, and there is even a script to make montages like this one using the low rez thumbnails(though you do have to be a bit careful with TOO many images or the total file size gets too large):
I am sure there are plenty of things I missed but if you are interested in more details, PM me (Red, I know you already did, just reply) and I can send you a discord link if you feel up to joining(again it's a YT'ers discord but we do discuss the seeds in there as we create them).
Red Eagle LXIX said:
I am quite keen to take that very top seed name and see just how good it is via Dan's rating system. I don't have any hard numbers for 6k seeds, so am not sure bad or well it would compare, but the image grouping of cities in the top right quadrant is particularly promising.