TL DR: I’m making a universal loot farming calculator that can answer the questions: what’s the best MP to farm for items, what’s the best route to farm for items and should you be opening chests to improve your farming efficiency? I’m coming to Reddit to get help collecting farming data from reliable individuals to create a model for this. PM me if you’re interested in helping out.
Hey guys,
You may have seen my reddit post the other day about this, but now I’m coming to you guys here at Diablofans to ask for help in collecting data so that we can produce a universal tool that can help us determine farming efficiency in terms of MFing. As many of you have figured out, farming at lower MPs with fast efficient builds is the best way to paragon level; however, since we have no correlate of XP/hr for loot-worthy items/hr, there is no concrete evidence that lower MPs are also the best for that task.
Moldran has recently done a 12 hour farming competition between him, Avoid and Invis to shed some light on the issue. Moldran was farming MP5 whereas the other two farmed MP 10. The only variable that they were counting was legendary drops. From the data I ran some simple statistical analysis to determine the effectiveness of such a measure.
The final results were Moldran: 44, Avoid 34 and Invis 45, which makes it seem like Moldran and Invis were much more efficient than Avoid; however, the t-tests when dividing the data up into Legendaries/hr only gave p-values of 0.22 and 0.19 respectively. When comparing Moldran and Invis against each other, the p-value jumped to 0.91. In post-analysis I went and tried to estimate the necessary N needed to obtain a p-value of 0.05 with a power of 0.8 using the pooled variance. For Moldran v Invis, they’d need 20 000+ hrs of farming data; for Invis v Avoid they’d need 167 hrs.
From this example it becomes clear that although theoretically Legendaries/hr is a good measurement for farming efficiency due to its linear scaling with MF and items/hr, it’s not very practical for answering questions and optimizing our runs.
The model that I’m planning on creating is one where certain rare drops are used as predictors for Legendaries/hr to decrease the RNG using the law of large numbers. By that I mean, if we knew the ratio of rare 58+ gloves to legendaries was 4:1, when a farming run turns up with 8 gloves/hr its equivalent to 2 legendaries/hr which gives a secondary measurement of legendaries/hr. With more estimates the RNG induced variability is reduced. I’m also considering adding additional weight to rares based on their expected population variance.
Those of you familiar with the nuances of how magic find mechanics work in the game (posted by the wonderful group here at Diablofans) are probably aware that these ratios are dependent on the MF of the player. Their research has shown that the quantity of rare drops has diminishing returns with magic find due to the sequential roll theory, whereas legendaries have a (as far as we know) direct correlation.
Over the past 2 weeks I (and other trusted individuals) have collected some preliminary empirical data from our farming runs, only keeping track of our 58+ gloves, amulets, rings and legendaries. Using our pitifully small sample size of 100 legendaries collected I decided to just split the results into MF of 400-575 and MF of 189-400 for now. What I found is that the ratio of gloves rings and amulets to legendaries for the 400+ group was 3.47, 1.89 and 1.04 respectively; for the 400- MF group it was 4.26, 2.35 and 1.17. This amounts to a reduction of 18.6%, 19.3% and 11% in the three groups demonstrating that indeed the ratio of rares:legendaries will decrease with increasing MF (p of 0.003 by ANOVA).
Now that I’m confident that the empirical data matches the theoretical MF analysis, I’m planning on opening up this project for others to help in the data collection. What I hope to do is be able to determine a formula to calculate the ratio that incorporates the user’s MF. At the moment I’m unsure whether MP bonus drops will have a measureable effect on this ratio, so I’m going to be playing safe and collected that tid bit of data in case I have run a regression analysis for that.
So what data will I be collecting? Stuff that people normally are picking up: 58+ gloves, 58+ rings, 58+ amulets, 61 1-handers, 62 1-handers, 63 1-handers (all excluding 1 handed xbows), 63 shoulders, 63 pants, 58+ bracers meaning you would generate 10 estimates of Leg/hr. Once I’m happy with the reliability of the model I’ll release a beta version for the public to try out. This will provide a method for people to figure out which MP they should be farming items on, the theorycrafters out there to determine the best routes for farming items as well as a way to measure our epeen against others.
If you are interested in helping me collect data, please send me a pm or a twitch pm at chazhangarang (I prefer this since the Twitch banner is a much more innocuous looking email to open at work). To avoid trolls messing up the data, I’m going to be restricting the data submission to “reliable” people that I feel like I can trust based off of their post history or history in Diablo in general. I could really use some data from the top high MP farmers that browse these forums.
The offline sheet can be found here (note that run times should be measured in min using decimals). I'll make an update when the online form will open up this weekend. Again, pm me here or preferably via twitch and I'll try to get to you as soon as possible. You should also add in notes for your runs and save your data so that in the future when I release the calculator you can have data ready to input. Here's a FAQ for the sheet:
Is there a specific route to run? The only rule is that it must be MP0 Act 3 and above to maintain the same ilvl drop ratios. Any route is acceptable as long as it gets 5 NV stacks.
What MF should I put in? You should enter your base MF. Pretty much the MF that would show up in your details panel on MP0 with 0 NV stacks.
How should I deal with 5 NV stacks affecting my MF? There are 2 options. 1. Time yourself for 3 runs to determine your run length and time to obtain 5 stacks and enter that data into the chart. This data can be reused if your runtime doesn't change. 2. Record your drops pre 5 NV and post 5 NV and submit it as 2 different data sets. The first set with a run time of 1 and 5 NV time of 1 and the 2nd as a run time of 1 and 5 NV time of 0.
How should I group the data? Data should be grouped based on the MP that you ran, the specific route you ran and your MF. For MF there's a bit of wiggle room of up to 6% to account for your paragon level gains. ie. If you did an Alk run at MP0 and MP 5 they would be different sets. If you tried a full Act 3 clear and Alk run at MP0, they would be different sets (unless you're using the 2nd method of dealing with 5 NV listed above). If you were doing an Alk run on MP1 continuously from Paragon lvl 50-52 and went from 280 to 286% mf you can feel free to lump those together. Once you get to paragon lvl 53 you would make a new data set. I'm going to be opening up a google doc submission once or twice a week, so you can lump all of your data together using the above rules until the submission time.
Whats the minimum data set size? I'm going to be limiting data sets to be 2 or more legendaries. Although I realize this will make it so that people with 100 or less MF will unlikely be able to submit data (which also means the tool won't be useful for predicting results in that range), the tool is meant to be used to increase MFing efficiency. If you have less than 100 MF then the solution would be to get more MF rather than use this tool.
Do I have to loot everything listed?
Ideally yes. The more loot people are collecting the more accurate the resulting data model will be. If it really is a game breaker for participating, then you can just loot what you normally loot and write down a comment indicating what you item slots you didn't loot.
Thanks for putting up with my ramblings in this convoluted post!
would it be easier to video record the runs and then go back and pinpoint the data needed from videos? (for the people that can do this) or is it just as easy to stop during runs to jot down data. i would think recording the runs and then getting the info would have to be more efficient data gathering since the runs would be normalized from not having to stop every few minutes to write down stuff.
Hey guys,
You may have seen my reddit post the other day about this, but now I’m coming to you guys here at Diablofans to ask for help in collecting data so that we can produce a universal tool that can help us determine farming efficiency in terms of MFing. As many of you have figured out, farming at lower MPs with fast efficient builds is the best way to paragon level; however, since we have no correlate of XP/hr for loot-worthy items/hr, there is no concrete evidence that lower MPs are also the best for that task.
Moldran has recently done a 12 hour farming competition between him, Avoid and Invis to shed some light on the issue. Moldran was farming MP5 whereas the other two farmed MP 10. The only variable that they were counting was legendary drops. From the data I ran some simple statistical analysis to determine the effectiveness of such a measure.
The final results were Moldran: 44, Avoid 34 and Invis 45, which makes it seem like Moldran and Invis were much more efficient than Avoid; however, the t-tests when dividing the data up into Legendaries/hr only gave p-values of 0.22 and 0.19 respectively. When comparing Moldran and Invis against each other, the p-value jumped to 0.91. In post-analysis I went and tried to estimate the necessary N needed to obtain a p-value of 0.05 with a power of 0.8 using the pooled variance. For Moldran v Invis, they’d need 20 000+ hrs of farming data; for Invis v Avoid they’d need 167 hrs.
From this example it becomes clear that although theoretically Legendaries/hr is a good measurement for farming efficiency due to its linear scaling with MF and items/hr, it’s not very practical for answering questions and optimizing our runs.
The model that I’m planning on creating is one where certain rare drops are used as predictors for Legendaries/hr to decrease the RNG using the law of large numbers. By that I mean, if we knew the ratio of rare 58+ gloves to legendaries was 4:1, when a farming run turns up with 8 gloves/hr its equivalent to 2 legendaries/hr which gives a secondary measurement of legendaries/hr. With more estimates the RNG induced variability is reduced. I’m also considering adding additional weight to rares based on their expected population variance.
Those of you familiar with the nuances of how magic find mechanics work in the game (posted by the wonderful group here at Diablofans) are probably aware that these ratios are dependent on the MF of the player. Their research has shown that the quantity of rare drops has diminishing returns with magic find due to the sequential roll theory, whereas legendaries have a (as far as we know) direct correlation.
Over the past 2 weeks I (and other trusted individuals) have collected some preliminary empirical data from our farming runs, only keeping track of our 58+ gloves, amulets, rings and legendaries. Using our pitifully small sample size of 100 legendaries collected I decided to just split the results into MF of 400-575 and MF of 189-400 for now. What I found is that the ratio of gloves rings and amulets to legendaries for the 400+ group was 3.47, 1.89 and 1.04 respectively; for the 400- MF group it was 4.26, 2.35 and 1.17. This amounts to a reduction of 18.6%, 19.3% and 11% in the three groups demonstrating that indeed the ratio of rares:legendaries will decrease with increasing MF (p of 0.003 by ANOVA).
Now that I’m confident that the empirical data matches the theoretical MF analysis, I’m planning on opening up this project for others to help in the data collection. What I hope to do is be able to determine a formula to calculate the ratio that incorporates the user’s MF. At the moment I’m unsure whether MP bonus drops will have a measureable effect on this ratio, so I’m going to be playing safe and collected that tid bit of data in case I have run a regression analysis for that.
So what data will I be collecting? Stuff that people normally are picking up: 58+ gloves, 58+ rings, 58+ amulets, 61 1-handers, 62 1-handers, 63 1-handers (all excluding 1 handed xbows), 63 shoulders, 63 pants, 58+ bracers meaning you would generate 10 estimates of Leg/hr.
Once I’m happy with the reliability of the model I’ll release a beta version for the public to try out. This will provide a method for people to figure out which MP they should be farming items on, the theorycrafters out there to determine the best routes for farming items as well as a way to measure our epeen against others.
If you are interested in helping me collect data, please send me a pm or a twitch pm at chazhangarang (I prefer this since the Twitch banner is a much more innocuous looking email to open at work). To avoid trolls messing up the data, I’m going to be restricting the data submission to “reliable” people that I feel like I can trust based off of their post history or history in Diablo in general. I could really use some data from the top high MP farmers that browse these forums.
The offline sheet can be found here (note that run times should be measured in min using decimals). I'll make an update when the online form will open up this weekend. Again, pm me here or preferably via twitch and I'll try to get to you as soon as possible. You should also add in notes for your runs and save your data so that in the future when I release the calculator you can have data ready to input.
Here's a FAQ for the sheet:
Is there a specific route to run?
The only rule is that it must be MP0 Act 3 and above to maintain the same ilvl drop ratios. Any route is acceptable as long as it gets 5 NV stacks.
What MF should I put in?
You should enter your base MF. Pretty much the MF that would show up in your details panel on MP0 with 0 NV stacks.
How should I deal with 5 NV stacks affecting my MF?
There are 2 options. 1. Time yourself for 3 runs to determine your run length and time to obtain 5 stacks and enter that data into the chart. This data can be reused if your runtime doesn't change. 2. Record your drops pre 5 NV and post 5 NV and submit it as 2 different data sets. The first set with a run time of 1 and 5 NV time of 1 and the 2nd as a run time of 1 and 5 NV time of 0.
How should I group the data?
Data should be grouped based on the MP that you ran, the specific route you ran and your MF. For MF there's a bit of wiggle room of up to 6% to account for your paragon level gains. ie. If you did an Alk run at MP0 and MP 5 they would be different sets. If you tried a full Act 3 clear and Alk run at MP0, they would be different sets (unless you're using the 2nd method of dealing with 5 NV listed above). If you were doing an Alk run on MP1 continuously from Paragon lvl 50-52 and went from 280 to 286% mf you can feel free to lump those together. Once you get to paragon lvl 53 you would make a new data set.
I'm going to be opening up a google doc submission once or twice a week, so you can lump all of your data together using the above rules until the submission time.
Whats the minimum data set size?
I'm going to be limiting data sets to be 2 or more legendaries. Although I realize this will make it so that people with 100 or less MF will unlikely be able to submit data (which also means the tool won't be useful for predicting results in that range), the tool is meant to be used to increase MFing efficiency. If you have less than 100 MF then the solution would be to get more MF rather than use this tool.
Do I have to loot everything listed?
Ideally yes. The more loot people are collecting the more accurate the resulting data model will be. If it really is a game breaker for participating, then you can just loot what you normally loot and write down a comment indicating what you item slots you didn't loot.
Thanks for putting up with my ramblings in this convoluted post!
sounds like a good idea but a ton of work. lol
http://us.battle.net...22/hero/3153312 (wiz)
http://us.battle.net/d3/en/profile/heavymetal-1322/hero/8654665 (monk)
my twitch feed (just messing around, no super player or anything)
http://www.justin.tv/heavymetalmak