Statistical Map Research

I added some quick facts sheet summarizing some easy to read not too mathematical info before open beta launches tomorrow.


Experience

Out of the 413 map runs, the average map gives 4,343,890 exp

Maze maps usually give 2,515,598 additional exp
Area size maps usually give 29,607 additional exp per 1% larger area (Maze similar exp to 85% larger area)
Larger pack size maps usually give 8,130 additional exp per 1% pack size


Map drops

Out of the 413 map runs, 457 maps dropped in total
That is an average of 1.11 map drops per map run

A map with a 0% item quantity bonus would on average drop 0.80 maps
Each +1% quantity bonus would add 0.0107 maps on top of that
It takes 93% more map item quantity bonus to get 1 more map dropped on average



Read this data carefully as it's based on a relatively small data sample and on some limiting assumptions.

I'm still looking for contributors for more data when open beta launches and for help on statistical theories and models (multiple regression & anova models, principal components analysis etc).

Sheet url:
https://docs.google.com/spreadsheet/ccc?key=0AvSHrzEQ02ZBdHIxcHBCUkZCRlhXWUNuVVNjeUJGa0E

Thanks and keep bashing!

Zoom have you started the data sheet for open beta yet? I'll give this thread a bump for those who are doing maps already and would like to contribute.

I'm still in merciless act 2. A ways from maps yet.

Pretty curious how, or if, this project is going so far in OB?

I havent reached map stage yet since I play on hardcore and my first char died to a disconnection. Leveling a second one now, hopefully I'll get there soon.

I'll probably restart a new datasheet from scratch.

Anyway, I'm ready to record data if you guys are allready running maps :)

Looks like you've already done a few maps. I copied a new sheet for open beta and am about to try a couple maps now. Still a bit under geared I think, so it will be slow going.

The link to my google doc is still the same.

EDIT: going pretty well so far. Map drop rates must have went down since CB- or I'm just getting some bad RNG. That or map drops are somehow effected by map level- perhaps larger loot pools, or some scaling factor that may not have been taken into account for the higher level range of maps in OB? Shrug. Hope it get's better or I'm gunna run out of maps.

I had started running some maps and copied yours FriarJon, thanks!
I also had some bad luck on drops apparently and I keep running out.

Anyway, it looks like 0.10.1 changed maze affix and modified some maps so I'll just discard the 13 runs we had so far.

If I may make a suggestion, map drops will be affected by all the following things:
- Current map's level
- Current map's quantity bonus
- Current map's demographics

The third bit you're passing over. You're poised to see a result like: "What kind of map drops do I get from a cells map?" - This will include level breakdowns, what percent of them are level 72 maps, 71 maps, .... etc. And some sort of correction for this data to 0% quant, then a trend against quant. All great stuff so far.

But now, which demographic map mods you happen to favor will skew your results. You have to do the entire exercise five times:
- Once for all maps with Pack Size
- Once for all maps with Maze / Area
- Once for all maps with Magic Mobs
- Once for all maps with Rare Mobs
- Once for all maps with Twinned Boss.
... Actually, six, because you will want information on the base demographic as well.

... And you have to neglect results for any maps that overlap multiple demographic mods. All 5 of those mods drastically change a map's drop profile.

I assume that you want to know which combinations of map mods are best for forward map progression. The raw information you need to compute from all this goes something like: 0% quant odds of getting a map to drop across the whole vector of possible levels. But, that information only makes sense in context with the mob demographics. A default mob population will have lower odds and lower on the level gradient. With increased pack size you will stay in the same place on the level gradient but your probabilities will improve. With a twinned boss you will have higher than average probabilities and also land higher on the level gradient.

This is the level of detail you need to actually make predictions about how other more complicated mod combos will interact. But, there's a problem: now you need literally ten thousand datapoints to trend anything. And the data collection is problematic because you can't just make it a community log project. Far too complicated, such that a large enough percentage of data submissions will be flawed and poison the results. You could only rely on input from a small group of trusted friends.

This is all why I thought the idea over some time ago and didn't touch it.

Hi and thanks for your suggestions!

After reading your post several times though, I'm not entirely sure to understand your points (english isnt my primary language)

First let me tell you that I record all the "demographics" data you mentionned. I actually do a univariant linear regression of xp and mapdrops versus all the variables you mentionned and look at what correlations look significative. The univariant models are obviously not optimal but I havent found a way to work out multiple regressions easily in a google doc spreadsheet. I think univariant models are good enough if we assume all variables are independant.
This is obviously not the case, namely because most "demographics" mods have a negative correlation on item quantity bonus. That's a problem that will bias the conclusions but I'm aware of it and trying to work out a solution, whether by flattening, working on different indicators like maybe drop/expgain...

I also prefer to not draw any conclusions too early, I'm perfectly aware it will take thousands of map runs to be able to reliably read results. I'm just having fun doing it and it only takes me 1 minute or so to write down the exp and map details after each map run.

At the moment, I still need more contributors because out of the 450 maps we had in the closed beta spreadsheet, I had run like 350 or so myself and I'm probably not gonna be able to keep this rhythm in the next future. And I'm not buying the argument that it's too complicated, it seriously doesnt take more than 30s-1m per map to write down the required information. I'll initiate a post on Reddit or something in the next few days to get more people helping :)

Zaka, I'm sure Zoom would be grateful for any help or advice you can offer.