• 1

    posted a message on Magic Find and its efficiency: A statistical insight
    This is fantastic work so far. The data you've collected is a huge dataset, and I think there's a ton more that can be done to cut it in interesting ways. The graphs right now have a very funny shape ("I find fewer blue items with more MF?!") and it's because the data are represented in an overlayed graph... A stacked graph might be better (eg, there's a line that shows white items, a line that shows [white+blue], a line that shows [white+blue+yellow], and the regions between the lines are colored in appropriately.)

    Something to consider is that goblins (unlike trash mobs) have guaranteed loot drops: it looks like the goblins may have a fixed minimum of 1 blue and 2 white items, but these individual items have a chance to be scaled up. There's also a VERY high probability that one of the whites gets upped to a blue: in all the data, only 3 goblins ever dropped only 1 blue-or-better. all three happened to be "Seeker" type enemies.

    Based on the numbers I'm seeing, It's likely that these guaranteed values differ by goblin type, too (as seen in your previous research):
    Pygmys always drop at least 1 white, 3 blue
    Seekers always drop at least 2 whites, 1 blue (almost always 1 white gets upgraded though)
    Bandits always drop at least 2 whites, 2 blues (almost always 1 white gets upgraded though)
    Goblins always drop at least 1 white, 3 blues

    It would be interesting to look at the subset of cases where the goblins dropped exactly the minimum number of items, and they did drop a white item. This would help us see exactly how MF affects the chance of a blue item getting upgraded. From there, work your way up to seeing what it takes to upgrade that first white item, and then look at the cases where exactly one extra item dropped and see how MF affects just that one bonus item (I would guess that it's similar to how the first 'freebie' white item works).

    I'd love to spend more time cutting up this data, but I have to get back to work. Once again, thank you for your persistence in pulling in all this great data.
    Posted in: Theorycrafting and Analysis
  • To post a comment, please or register a new account.