Research on Research, or why putting 3 items instead of 5 is better long-term

[AstroOwl]

TL,DR: When you want to upgrade a lot of items to the next rarity and want on average to get more higher rarity items, you should research packs of 3 items, not 4 or 5.

 

Now, details, explanation, and why that is so.

 

The question i asked myself is: suppose i have a lot of items to upgrade. I don't need guarantees that i get next-rarity item on each craft. But i want to get maximum higher-rarity items out of a lot of lower-quality ones. Which way of research (3,4, or 5) should i use?

 

Let's suppose for clarity that you have a lot of Uncommon items and want to upgrade them to rare.

 

Let's now observe profit from possible options of investing Uncommon items. I would do so in two ways, for clarity and better understanding.

 

Way 1, harder to understand, but more strictly done:

 

 

 

At any time, you can convert 5 Uncommon to 1 Rare. So let's say: 1 Rare costs 5 Uncommon.

 

"Profit" is how much you get from research run in value of Uncommon items.

 

Option 1, put 5 items: I sacrifice 5 Uncommon and get 1 Rare, profit = 5 - 5 = 0.

 

Option 2, put 4 items:

With the chance of 80% you get 1 Rare, and your profit is 5-4 = +1. Weighted profit is (profit*probability)=(+1*0.8 )=0.8

With the chance of 20% you get back just an Uncommon item, and your profit is negative: 1 - 4 = -3. Weighted: (-3*0.2)=-0.6

And the total profit is 0.8 - 0.6 = 0.2.

That means that on average on each research you get 0.2 equivalent of an Uncommon item back.

 

Option 3, put 3 items:

With the chance of 60% i get 1 Rare, and your profit is 5-3 = +2. Weighted profit is (profit*probability)=(+2*0.6)=1.2

With the chance of 40% i get back just an Uncommon item, and your profit is negative: 1 - 3 = -2. Weighted: (-2*0.4)=-0.8

And the total profit is 1.2 - 0.8 = 0.4.

That means that on average on each research you get 0.4 equivalent of an Uncommon item back.

 

 

Way 2, easier to understand:

 

 

Let's say you have 120 Uncommon items (i pick large number intentionally, to show the fact that it is thing that works on average, that is, better with large numbers).

 

Option 1: You put them in in packs of 5. That makes enough for 24 research runs, and you with 100% chance get 24 rare items.

 

Option 2: You put them in in packs of 4. That makes enough for 30 research runs.

On average, 80% of them (30*0.8=24) would yield you a Rare item each. So, you get 24 Rare items.

Another 20% of them (6) would return to you an Uncommon item.

So, after that runs, you get 24 Rare + 6 uncommon on top of that!

 

Option 3: You put them in in packs of 3. That makes enough for 40 research runs.

On average, 60% of them (40*0.6=24) would yield you a Rare item each. So, you get 24 Rare items.

Another 20% of them (40*0.4=16) would return to you an Uncommon item.

So, after that runs, you get 24 Rare + 16 uncommon on top of that!

 

As can be clearly seen, when you invest them in smaller packs, you get the same amount of Rare items, but on top of that, you get some Uncommon items, which you can invest once again!

 

 

 

 

TL,DR version of the explanation: investing some amount of, say, Uncommon items in packs of 3 would yield you on average as much Rare items as you would get by investing in packs of 5, but on top of that, you would get ~13% of your Uncommon items back, and would be able to turn them in for additional Rares.

[Pob]

Interesting math

there's also another reason to do groups of 3 not 5 and that is types, I've not found the math for it however so far it seems that the object you get out depends on what you put in, so if I want a better armed turret module then the only way to guaranty it is to only put in armed turret modules.

Now turret upgrades are not the best example here because it's a fixed upgrade, however look at something like actual turrets, more chances to get the specific bonus you're looking for, esp with the additional bonuses like independent targeting, extra range or guided rockets.

 

Does anyone know the out come effects on mixing objects for upgrade? from my personal experience it seems like each type has an equal chance, so put in 4 chainguns and 1 cannon and you've got a 50% chance to get a chaingun and a 50% to get a cannon, not a 80% chance to get a chaingun and an 20% chance to get a cannon.

 

[JiggsLuker]

This is a very interesting thought process. I tried it a bit when I first started but without the math to back it up. I think I am going to give it another shot.

[Warrax]

Trouble is ... With any math ... When putting less than 5 in, you can get the same back, meaning no upgrade in rarity.

 

 

I dunno.  In my gameplay, i am not eager to make "more" money.  I am interested in getting Exotic and Legendary items as soon as possible.

To me, common, uncommon (etc) items are a direct path to the best items.

 

 

If i want money, i will scan for sellable asteroids, or do missions or trade.

 

I care about money only as far as it pays for my crew and can upgrade my ships.

 

 

[AstroOwl]

Trouble is ... With any math ... When putting less than 5 in, you can get the same back, meaning no upgrade in rarity.

 

Really? I was completely unaware of that (sarcasm)

 

Seriously, i did accounted for that. You don't even need to understand any math to see that i did.

 

If you want a version without any math whatsoever: When you put less than 5 in, you spend less items on each research run => you get more runs => even with reduced chances of success, more runs are enough to get the same amount of next rarity items as with 5-runs + some spare no-upgrade items return from unsuccessful runs.

 

 

[Blondersheel]

I dunno.  In my gameplay, i am not eager to make "more" money.  I am interested in getting Exotic and Legendary items as soon as possible.

I don't think research stations are a path to money any way you look at it...if you want money just sell the items. However, if you want exotic and legendary items, then AstroOwl showed the math that says its better do upgrade in groups of 3.

 

The only scenario where I would use 5 is when I have a group of 5 exotics and I want to be sure I get a legendary. Even though you're still better off over the long run to upgrade in groups of 3, it's really shitty to lose a bunch of exotics when you have a bad luck streak.

 

Here's another way to look at it: If you start with common items and upgrade them through research, how many will it take you to get a legendary with groups of 3 or 5?

With groups of 5: it will take you 3,125 common items (at 5 per upgrade)

With groups of 3: it will take on average 1,545 common items (at an average ~4.34 per upgrade)

 

That's almost exactly half the number of common items required to get a legendary! Clearly it is better to upgrade in groups of 3 when doing any kind of bulk upgrading.

 

[spurgurgle]

i see it like this

 

if you only have a few to research your better off doing it in 5's, just to play it safe

 

but if you have lots to research just do it in 3's

 

if you take 5 you can do a set of 3. if i works great you have your upgraded one and still have 2 spare...

if it didnt work you get 1 non upgraded one to add to your other 2 to do another group of 3.

with the same chance of getting a upgrade or not

[MagusUnion]

Did you notice any quality difference in the items that resulted in inputting 3 instead of 5? How does the decrease in the amount of items influence the stat averages for drops?

[SageThe13th]

Another advantage of running 3 item researches is that you are constantly rerolling the dice on the resulting item's stats.  This gives you change to get a good item even if it's at the same rarity level as the items you put in.  Also an item's sale price is based on it's stats.  I have a legendary that sadly ended up with patently average stats and it's worth far less than my exotics that all happen to have very good stats.  So giving the RNG more chances to split out random well stat'd items will make you more money than having fewer items of a guaranteed rarity level at the end of the day.

[AstroOwl]

Another advantage of running 3 item researches is that you are constantly rerolling the dice on the resulting item's stats. (...) I have a legendary that sadly ended up with patently average stats and it's worth far less than my exotics that all happen to have very good stats.  (...)

 

This. Not even talking in terms of money, but since the stats vary so widely, even for same rarity, and "higher rarity" don't guarantee better stats, it's very advantageous to roll often.

 

Did you notice any quality difference in the items that resulted in inputting 3 instead of 5? How does the decrease in the amount of items influence the stat averages for drops?

 

Haven't noticed any difference, from what i see, if you happen to get a, say, exoic item, game just says "good, now i roll for a random stats for exotic item, with exotic stat averages". Way of getting that exotic item - via loot, or research, or how many of lower-tier was invested, seem to not influence the stat averages.

 

if you only have a few to research your better off doing it in 5's, just to play it safe

 

but if you have lots to research just do it in 3's

 

basically, yes. If you can spare just one slot for that upgrade type, it's better certainly be one rarity higher.. although for some upgrades higher rarity do not guarantee better stats for you =)

 

Does anyone know the out come effects on mixing objects for upgrade? from my personal experience it seems like each type has an equal chance, so put in 4 chainguns and 1 cannon and you've got a 50% chance to get a chaingun and a 50% to get a cannon, not a 80% chance to get a chaingun and an 20% chance to get a cannon.

 

Can't say anything reliable on this particular matter, unfortunately.

 

Maybe i should check researchstation.lua and see if i find more details there.

[Pob]

Does anyone know the out come effects on mixing objects for upgrade? from my personal experience it seems like each type has an equal chance, so put in 4 chainguns and 1 cannon and you've got a 50% chance to get a chaingun and a 50% to get a cannon, not a 80% chance to get a chaingun and an 20% chance to get a cannon.

 

Can't say anything reliable on this particular matter, unfortunately.

 

Maybe i should check researchstation.lua and see if i find more details there.

it'd be nice to have my suspicions proved or find out if I have bad luck, I've taken to selling repair and force turrets to stop them making more and trade modules after the 3rd exotic

 

here's something else I noticed, if you mix upgrades with turrets then if you get a turret it can be made of iron even if none of the turrets where.

So I think all upgrades count as iron material.

 

Material is another one of those factors which seem to be even chance, regardless of how many of which type are put in.

ie put in 4 trinium and 1 iron and you have a 50/50 chance of getting an iron or trinium out

again someone needs to dig around in the code to confirm this

 

This means you should never mix upgrades with high tier material turrets.

[nasabot]

Its funny, because today I saw a YT video where the guy stated that he explicitly did the math and was sure that spending 5 is the best.

 

But now I did the math myself and its true:

 

3>4>5

 

 

My approach:

 

5:

You spend 5x 0,2 in order to get 1

(1 x 1 + 0 * 0,2) /5 = 0,2

0,2 => 0,2

 

4:

You get

(0,8 x 1 + 0,2 * 0,2) /4 = 0,21

0,2 => 0,21

 

3:

You get

(0,6 x 1 + 0,4 * 0,2) /3 = 0,227

0,2 => 0,227

 

Therefor you have 13%(0,227/0,2) more output if you use 3 instead 5.

[Pob]

It depends on what you want to get out of the research

If you have loads and you just want as many better rarity as possible or you don't need that upgrade right now, then yes 3 is better.

If however you are after specific types of turret/upgrade/material then you'll generally have far less of any one type so to guaranty a specific item of higher rarity you're better off using 5 of the same.

 

 

[artforz]

Its funny, because today I saw a YT video where the guy stated that he explicitly did the math and was sure that spending 5 is the best.

 

But now I did the math myself and its true:

 

3>4>5

 

 

My approach:

 

5:

You spend 5x 0,2 in order to get 1

(1 x 1 + 0 * 0,2) /5 = 0,2

0,2 => 0,2

 

4:

You get

(0,8 x 1 + 0,2 * 0,2) /4 = 0,21

0,2 => 0,21

 

3:

You get

(0,6 x 1 + 0,4 * 0,2) /3 = 0,227

0,2 => 0,227

 

Therefor you have 13%(0,227/0,2) more output if you use 3 instead 5.

I also did the math, but got a slightly different result:

 

My approach: "how many uncommon do you put in, how many uncommon and rare do you get out on 10 runs of 3/4/5 uncommon inputs with perfectly average luck"

 

5: 10*5 uncommon in, 10 win/0 loss so 10 rare out.

10 rare / 50 uncommon = 0.2 rare/uncommon

 

4: 10*4 uncommon in, 8 win/2 loss so 8 rare out, 2 uncommon out.

8 rare / (40 uncommon - 2 uncommon) = ~0.21053 rare/uncommon

 

3: 10*3 uncommon in, 6 win/4 loss so 6 rare out, 4 uncommon out.

6 rare / (30 uncommon - 4 uncommon) = ~0.23077 rare/uncommon

 

So I get (6/26)/(1/5) = 15/13 = ~ +15.4% rares/uncommon from running sets of 3 instead of 5, not the +13% you got.

 

It can't be +15.4% and +13%, so there's gotta be a mistake somewhere.

 

Anyone care to point out the thinko?

[Ohm is Futile]

I'd be really interested in people keeping track in terms of what you put in for weapons and the results to figure out the chances of getting say a chaingun instead of a cannon if you put in 4 chainguns and a cannon, for example.

 

Seems like a large scale sample would be best to figure that out because that has a large influence on what you might get for weapons and mining/salvaging lasers.

[Pob]

well digging around in the code is probably better ;)

[Ohm is Futile]

If that's possible? I read that Koonschi does not want to release the code because the game is deeply tied to its engine. Anyways, been ages since I seriously looked at code at all, if someone can find that out then nice. If Koonschi is willing to disclose the information that would be even nicer. For now though, gathering data is the best I can do.

[artforz]

If that's possible? I read that Koonschi does not want to release the code because the game is deeply tied to its engine. Anyways, been ages since I seriously looked at code at all, if someone can find that out then nice. If Koonschi is willing to disclose the information that would be even nicer. For now though, gathering data is the best I can do.

take a  look at data\scripts\entity\merchants\researchstation.lua ... other than rolling stats for the new module/turret, it's all done via scripting.

 

1. picks rarity using # and rarity of input items.

2. picks turret/module using input ratio of turrets/modules.

 

if it picked "turret":

3. picks weapon type using ratio of input turret weapon types

4. picks turret material using ratio of input turret weapon materials

5. picks "has independent targeting" using ratio of input turret having IT modifier.

6. generates a random turret of the selected material/rarity/weapontype, overrides IT modifier of result to what 5. rolled.

 

if it picked "module:

3. picks module type using input ratio of input module types

4. generates random module of the selected material/rarity/module type

 

TLDR: If you put in 4 chainguns and a cannon, you get chances of 80% chaingun and 20% cannon.

 

[Pob]

take a  look at data\scripts\entity\merchants\researchstation.lua ... other than rolling stats for the new module/turret, it's all done via scripting.

that's what I meant by looking at the code

 

ok looking at that script . . . my brain hurts

 

[Pob]

just looking at the weapon types

local weaponType = selectByWeight(random(), weaponTypes)

that is the line that specifies the type

the variable "weaponType" come from here

local weaponTypes = getWeaponProbabilities(items)

the variable "WeaponProbabilities" comes from here

function getWeaponTypesByIcon()

           

        if weaponTypes then

            return weaponTypes

        end

            weaponTypes = {}

       

           

        local weapons = Balancing_GetWeaponProbability(0, 0)

       

           

        for weaponType, _ in pairs(weapons) do

                   

            local turret = GenerateTurretTemplate(Seed(1), weaponType, 15, 5, Rarity(RarityType.Common), Material(MaterialType.Iron))

                    weaponTypes[turret.weaponIcon] = weaponType

               

        end

       

           

        return weaponTypes

    end

   

    function getWeaponProbabilities(items)

           

        local probabilities = {}

           

        local typesByIcons = getWeaponTypesByIcon()

       

           

        for _, item in pairs(items) do

                    if item.itemType == InventoryItemType.Turret

                            or item.itemType == InventoryItemType.TurretTemplate then

               

                           

                local weaponType = typesByIcons[item.weaponIcon]

                           

                local p = probabilities[weaponType] or 0

                            p = p + 1

                            probabilities[weaponType] = p

                       

            end

               

        end

       

           

        return probabilities

    end

and that's where I struggle.

it's using the Icon to create a value for the weapon type? ie chaingun = 1, bolter = 2, etc

But my limited understanding fails trying to understand how it's generating the probabilities  :o

[Ohm is Futile]

If that's possible? I read that Koonschi does not want to release the code because the game is deeply tied to its engine. Anyways, been ages since I seriously looked at code at all, if someone can find that out then nice. If Koonschi is willing to disclose the information that would be even nicer. For now though, gathering data is the best I can do.

take a  look at data\scripts\entity\merchants\researchstation.lua ... other than rolling stats for the new module/turret, it's all done via scripting.

 

1. picks rarity using # and rarity of input items.

2. picks turret/module using input ratio of turrets/modules.

 

if it picked "turret":

3. picks weapon type using ratio of input turret weapon types

4. picks turret material using ratio of input turret weapon materials

5. picks "has independent targeting" using ratio of input turret having IT modifier.

6. generates a random turret of the selected material/rarity/weapontype, overrides IT modifier of result to what 5. rolled.

 

if it picked "module:

3. picks module type using input ratio of input module types

4. generates random module of the selected material/rarity/module type

 

TLDR: If you put in 4 chainguns and a cannon, you get chances of 80% chaingun and 20% cannon.

Thanks, that corresponds to what's intuitive (imo, anyways) and answers my question. Perhaps more interesting is that the "quality" or rather the previous numbers a module/weapon had don't matter, all that matters is material, IT and type. The rest is completely randomized.

 

Did you see anything regarding tech level? My current assumption is that it only depends on how far or close the research station is from the center of the galaxy rather than the tech level of the individual items.

[artforz]

Sure looks like it.

There's noting done explicitly for TL, and following what's done with x,y from researchstation transform -> turretgenerator TurretGenerator.generate -> galaxy Balancing_Get...

dps and tech values passed into GenerateTurretTemplate are solely derived from floor(distance to center of research station's sector)-5