Loot Table Holes

[Zenji_]

Maybe it's anecdotal, but I swear, in some games, some items just plain never show up. Ever. It's not a huge problem for schematics that can be learned via talents, but otherwise, you're left having to resort to the creative menu if you can. That's not possible on a multiplayer server.

 

For example, in my latest game, I'm up to day 59. I have visited at least 3 traders every time the inventories turn. I have done numerous POIs, including tier 5s. I am level 51.

 

I have never seen a silencer, or a silencer schematic. Not once. Nor have I seen the meat stew recipe. It seems I've seen most other things several times.

 

Why is this? Is this by design or a bug? It strikes me as a bug. I know RNG can be horrible sometimes, but this is ridiculous.

 

Have the devs done test runs to verify that their loot generation algorithm is correctly covering their entire loot table?

[myrkana]

Have you though that a silencer is a very high level of loot then? Maybe you're still not high enough for it, or not getting lucky? Luck does still play a large part in your loot

[Zenji_]

No. I have a flawless desert eagle. A silencer isn't that amazing, unless they changed its rarity for A19. In other 7d2d play-thrus I have played, the silencer is readily available practically always, in the early mid-game. I'm at the end game now. I fear nothing. And yet, no silencers. If indeed they did make it some kind of super-ultra rare end game item, that would be ....interesting. I'd disagree with that change. It'd make POIs not harder, just more tedious, because I'd resort to a crossbow. With their extra +200% sneak damage in A19, they are devastating. 

 

And again, it's not just the silencer. It's other items as well. Are you telling me a meat stew is end game amazingness? This also is't new in A19. In previous games I've played, stuff is mysteriously missing from the loot table. For example, pumpkin seeds. Some games I've played, I never ever saw them. Why?

[myrkana]

8 minutes ago, Zenji_ said:

No. I have a flawless desert eagle. A silencer isn't that amazing, unless they changed its rarity for A19. In other 7d2d play-thrus I have played, the silencer is readily available practically always, in the early mid-game. I'm at the end game now. I fear nothing. And yet, no silencers. If indeed they did make it some kind of super-ultra rare end game item, that would be ....interesting. I'd disagree with that change. It'd make POIs not harder, just more tedious, because I'd resort to a crossbow. With their extra +200% sneak damage in A19, they are devastating. 

 

And again, it's not just the silencer. It's other items as well. Are you telling me a meat stew is end game amazingness? This also is't new in A19. In previous games I've played, stuff is mysteriously missing from the loot table. For example, pumpkin seeds. Some games I've played, I never ever saw them. Why?

thats not missing from the table. Thats called crappy luck. Sometimes you dont see certain things sometimes you do.

 

I have not seen a silencer yet either, so might be an oversight. But the rest is just luck.

[Zenji_]

Another example: The semi mod. Never seen it in my current game. I've seen burst and full auto many times.

 

Sure, you can chalk it up to luck, but at some point, that become statistically very unlikely. I just want the Devs to do a pass on their algorithm. It's not a hard thing to do. Generate an appropriate sample of games, and run a test where loot is generated. Check for holes. It's possible there is an issue. It should not be dismissed because "luck lol".

[ElCabong]

This is my pet peeve about the game. It's always been like this. It's also one of the reasons I play in creative. I haven't seen a silencer drop yet. It's not very common but I saw them in A18 or the schematic.

 

Last game I played I got a 2x scope and then an 8X scope by day 10, haven't seen either one this time.

 

And then you really need to find a working workbench. Traders always sell them them eventually, very long after you're going to need one.

[CyrusBlaze]

6 minutes ago, Zenji_ said:

Another example: The semi mod. Never seen it in my current game. I've seen burst and full auto many times.

 

Sure, you can chalk it up to luck, but at some point, that become statistically very unlikely. I just want the Devs to do a pass on their algorithm. It's not a hard thing to do. Generate an appropriate sample of games, and run a test where loot is generated. Check for holes. It's possible there is an issue. It should not be dismissed because "luck lol".

 

At some point it might be statistically unlikely, but that's only after hundreds if not thousands of rolls for the exact loot group. With your sample size it absolutely can be chalked up to 'luck lol'


That being said here are the groups that control the gun part loot tables: (you can view the entire loot table via the loot file in the data/config folder if you want to see what is where and report an item actually missing)

 

<lootgroup name="modGunT1">
	<item name="modGunBipod"/>
	<item name="modGunChoke"/>
	<item name="modGunDuckbill"/>
	<item name="modGunFlashlight"/>
	<item name="modGunForegrip"/>
	<item name="modGunMagazineExtender"/>
	<item name="modGunMeleeTheHunter"/>
	<item name="modGunMuzzleBrake"/>
	<item name="modGunScopeSmall"/>
	<item name="modGunTriggerGroupSemi"/>
	<item name="modShotgunSawedOffBarrel"/>
</lootgroup>

<lootgroup name="modGunT2">
	<item name="modGunBarrelExtender"/>
	<item name="modGunCrippleEm"/>
	<item name="modGunBowArrowRest"/>
	<item name="modGunLaserSight"/>
	<item name="modGunReflexSight"/>
	<item name="modGunRetractingStock"/>
	<item name="modGunScopeMedium"/>
	<item name="modGunSoundSuppressorSilencer"/>
	<item name="modGunTriggerGroupBurst3"/>
	<item name="modGunShotgunTubeExtenderMagazine"/>
</lootgroup>

<lootgroup name="modGunT3">
	<item name="modGunBowPolymerString"/>
	<item name="modGunScopeLarge"/>
	<item name="modGunTriggerGroupAutomatic"/>
	<item name="modGunDrumMagazineExtender" prob=".3"/>
</lootgroup>

The Semi mod is actually a T1 mod, so with a higher gamestage the other tiers are also being rolled on reducing the chances of seeing them the further into progression you get (not impossible, but you have a higher chance at the lower tiers)

The Silencer is a T2 and in the same group as the burst mod you have gotten several of, so just a matter of the RNG gods not giving you what you want.

[Zenji_]

Do you know how many thousands of loot rolls have occurred in my game already, from POI chests, across 3 traders, numerous quest rewards, and zombie kills? How many thousands more do I need to log before you acknowledge that there may be an issue?

 

Look, what I'm wanting to see isn't difficult. It's standard for writing code. In fact, the devs (hopefully) are doing it already. I am just poking to make sure they are, because this is a negative experience with their game.

 

I want them to regression test the loot generation algorithm. Do it for however many iterations is necessary for statistical significance. Check to ensure they didn't make a mistake with it, that there isn't some kind of pathological corner case, or that it's behaving in a way they do not expect. It's entirely possible. And if all is well, then fine. All that's left after that is the nature of RNG. But based on what I'm seeing, they should want to make sure.

[Gr.o.m.]

You don't need the silencer mod to succesfully raid POIs or survive the horde night. So why is it so dramatically important that you have one?

[Boidster]

11 minutes ago, Zenji_ said:

I want them to regression test the loot generation algorithm. Do it for however many iterations is necessary for statistical significance. Check to ensure they didn't make a mistake with it, that there isn't some kind of pathological corner case, or that it's behaving in a way they do not expect. It's entirely possible. And if all is well, then fine. All that's left after that is the nature of RNG. But based on what I'm seeing, they should want to make sure.

They undoubtedly have, but what do you want from them? Like, an e-mail confirming "Yes, Zenji, we know how to do our jobs, here are the results from our regression test."?

 

The loot tables are there, in plain text, for you to read. You are apparently a skilled coder so interpreting them will not be difficult for you. They have even given you a peek into the model they used to produce the probabilities. The mods you are looking for are in the loot tables. You have received other loot from those same tables (given to you by RNG). Are you claiming that the algorithm somehow looks at that loot table and has a hard-coded error that excludes only the semi-auto mod from T1? How then would you explain other people who get the semi-auto mod?

 

I don't deny the frustration for RNG not giving you the thing you need - we all experience it. What we should be asking for, IMO, is a weighting/rubber-banding system which very gently increases the chance for a loot drop you've never had and (relatively speaking) decreases the chance for duplicates. IMO it should be very gentle indeed, not something that automatically gives you the semi-auto mod (for example) because that's the last one to Collect 'Em All. But if it bumps the chance for a semi-auto mod by 0.5% up to a maximum of +10% (pick whatever numbers make sense and/or are balanced) every time you don't get the mod in appropriate loot, that would help with fighting the RNG demon. I do not think that is on any of TFP's to-do lists, though, even the stretch-goal ones.

52 minutes ago, CyrusBlaze said:

The Silencer is a T2 and in the same group as the burst mod you have gotten several of, so just a matter of the RNG gods not giving you what you want.

Exactly correct. The loot tables are working, but RNG is being very churlish to Zenji. I sympathize. I was on day 60-ish on my A18 world and never did find a tablesaw schematic (and did not buy into the perk). Had to keep using the cabinet store's.

[JCrook1028]

You said it yourself, it's anecdotal. Period. End of discussion.

[seagas]

I found one, around day 65, but have yet to locate a Dessert Vulture on day 75. I started Horde every night on day 71 so five horde nights in a row, with Demos and no Vuture yet. Some times RNG is not in your favor.

 

[Gazz]

Random is random.

When hearing random, a lot of people expect something like a deck of cards that shuffles the order but still gets you every result in 30 or 60 rolls.

Random does not work like that at all.

There are hundreds of books and schematics to roll for.

12 hours ago, Boidster said:

What we should be asking for, IMO, is a weighting/rubber-banding system which very gently increases the chance for a loot drop you've never had and (relatively speaking) decreases the chance for duplicates.

That's a really complex system and unlike some MMOs we don't have any named armor sets where you must collect 5 specific rare drops.

 

[Ralathar44]

34 minutes ago, Gazz said:

Random is random.

When hearing random, a lot of people expect something like a deck of cards that shuffles the order but still gets you every result in 30 or 60 rolls.

Random does not work like that at all.

There are hundreds of books and schematics to roll for.

The loot tables having so much stuff these days is prolly driving down the chances of each individual item somewhat.  An embarrassment of riches has it's downsides.   IMO the reliable way to eventually find loot should be traders.  IIRC Traders are focused around having specific sorts of loot right?  So to cover our bases we'd need a specific trader who specialized in each other the following:

- Weapons and mods and schematics for such
- Tools and mods and schematics for such
- Resources/Vehicles/buildings (like forges/workbenches) and schematics for such
- Food and Medical and schematics for such
- All Schematics and perk books (double covers people looking for specific stuff, so they technically get 2 trader chances per 3 days)


I'm not sure what the current implemented or planned list of trader focus is.  But when implemented it should cover all the bases if possible and then have who focuses on what listed in the game's journal help system. (if it isn't already).  With trader quests eventually leading you to new traders you'd eventually be able to target what you want via traders.

[myrkana]

2 hours ago, Ralathar44 said:

The loot tables having so much stuff these days is prolly driving down the chances of each individual item somewhat.  An embarrassment of riches has it's downsides.   IMO the reliable way to eventually find loot should be traders.  IIRC Traders are focused around having specific sorts of loot right?  So to cover our bases we'd need a specific trader who specialized in each other the following:

- Weapons and mods and schematics for such
- Tools and mods and schematics for such
- Resources/Vehicles/buildings (like forges/workbenches) and schematics for such
- Food and Medical and schematics for such
- All Schematics and perk books (double covers people looking for specific stuff, so they technically get 2 trader chances per 3 days)


I'm not sure what the current implemented or planned list of trader focus is.  But when implemented it should cover all the bases if possible and then have who focuses on what listed in the game's journal help system. (if it isn't already).  With trader quests eventually leading you to new traders you'd eventually be able to target what you want via traders.

it would be cool is different traders specialised in certain things like Jen has more medical and base defence stuff, Joel has more ammo and weapons, etc... But then theyd also have to add a system on map spawn to make sure every trader is spawned? Or just leave it to randomness xD

[Boidster]

3 hours ago, Gazz said:

That's a really complex system and unlike some MMOs we don't have any named armor sets where you must collect 5 specific rare drops.

Hey, I'm just the idea guy; other people handle the details! 🙂

 

To be clear on this next part, I am not trying to be argumentative or to question the expertise and experience of you or any other TFP team member. More of a thought experiment and 'fun' thing to noodle out for a data engineer and former coder.

 

Here was my idea that the details interns have to accomplish:

 

Spoiler

 

• Add a new tag, "PraiseRNGesus" to items in items.xml that we want to nudge higher in probability the longer they go without being seen
• On initial game load, the items tagged PraiseRNGesus are put into a lookup table along with columns for FirstCheckedDay* and LastHitDay; this table is saved along with other game state information, and on subsequent loads items.xml is scanned and the table adjusted if needed (e.g. if the player added/removed the PraiseRNGesus tag to/from items)
• (This is the point where an ASSumption bites me in the @%$*#!) When loot is opened, all relevant loot groups are expanded and probabilities are calculated, I ASSume. I also assume that the result of that work is conceptually akin to an ordered list of items with an "if RNG result is less than X" (or maybe "between X and Y") criteria. RNG produces a number and the appropriate item is selected.
  • If loot groups are selected by RNG first, and only then are item-level probabilities calculated, this still should work
• I also assume - educated assumption perhaps - that the loot probabilities work by assigning a probability of "1" to all items in a group unless the item has a specific probability listed. Add them all up, then divide by the item's probability to get its true percent/floating point chance. So 5 items all with default probability (no specific "prob" tag in the XML) will have 20% chance each. If one of the 5 says prob=0.5, then the total of the 5 items will be 4.5 and the four default items would have 1/4.5 = 22% chance and the one special item would be 0.5/4.5 = 11%.
• The big payoff: when the calculations in the previous bullet are performed, the lookup table is checked and a very small nudge - perhaps 0.05, changeable in serverconfig - is given to items which have not been looted, weighted by the # of days since they were last looted. Assume the nudge is set for every 10 days. For each item (prior to calculating final probabilities):
  • If FirstCheckedDay is 0, this is the first time this item has ever been in a loot table, set FirstCheckedDay to game day and do not nudge
  • If FirstCheckedDay > 0 then nudge the item's prob by floor((Game Day - LastHitDay) / 10) * 0.05
  • For the item that hits, set its LastHitDay to game day; ixnay on the udgenay for at least 10 game days

 

And here is a walkthrough using a very tiny fake loot group:

 

Spoiler

<lootgroup name="groupFoobar">
    <item name="armorMilitaryHelmet"/>
    <item name="armorMilitaryBoots"/>
    <item name="armorSwatHelmet" prob=".4"/>
</lootgroup>

The denominator of the probability calc for this group is 1+1+.4 = 2.4, and the items have default probability of 41.7%, 41.7%, and 16.6% respectively (some rounding fudging here).

 

Assume all of those items have the "praiseRNGesus" tag on them in items.xml, like so:

<item name="armorMilitaryHelmet">
    <property name="Tags" value="priaseRNGesus,...other tags..."/>

 

1. Initial game load: items.xml is scanned and the praiseRNGesus items are pulled into a lookup table.

 

itemName	firstCheckedDay	lastHitDay
armorMilitaryHelmet	0	0
armorMilitaryBoots	0	0
armorSwatHelmet	0	0

 

2. Days 1-10: player doesn't open any containers that include lootgroup "Foobar"; no change to lookup table

 

3. Day 11: player opens a container and lootgroup "Foobar" is one of the possible results

• Algorithm checks all items in lookup table
• All items have firstCheckedDay=0, so firstCheckedDay gets set to 11* and no nudging of probabilities
• Probabilities:
  • Total denominator is 1.0 + 1.0 + 0.4 = 2.4 (same as default; no nudging)
  • Probabilities are 41.7%, 41.7%, and 16.6% (same as default)

Assume the player gets armorMilitaryHelmet.

• lastHitDay for armorMilitaryHelmet is set to 11

itemName	firstCheckedDay	lastHitDay
armorMilitaryHelmet	11	11
armorMilitaryBoots	11	0
armorSwatHelmet	11	0

 

4. Day 15: player opens another container for lootgroup Foobar

• All items have firstCheckedDay > 0 so we can check for the nudge
  • armorMilitaryHelmet (last hit on day 11): floor((gameDay - lastHitDay) / 10) = floor((15 - 11)/10) = floor (0.4) = 0
  • armorMilitaryBoots, armorSwatHelmet (never hit): floor((15 - 0)/10) = floor(1.5) = 1
• armorMilitaryHelmet gets no nudge
• armorMilitaryBoots and armorSwatHelmet get 1x the nudge factor of 0.05 added to their "prob" attribute
• Probabilities:
  • Total denominator is 1.0 (helmet) + 1.05 (boots) + 0.45 (swat helmet) = 2.50
  • armorMilitaryHelmet = 1.0/2.50 = 40%
  • armorMilitaryBoots = 1.05/2.50 = 42%
  • armorSwatHelmet = 0.45/2.50 = 18%

Assume the player gets armorMilitaryHelmet again in this loot. Presumably he comes here to the forums to complain. Meanwhile:

• lastHitDay for armorMilitaryHelmet is set to 15

itemName	firstCheckedDay	lastHitDay
armorMilitaryHelmet	11	15
armorMilitaryBoots	11	0
armorSwatHelmet	11	0

 

5. Day 20: player loots another container with lootgroup Foobar

• Check for the nudge
  • armorMilitaryHelmet (last hit on day 15): floor((gameDay - lastHitDay) / 10) = floor((20 - 15)/10) = floor (0.5) = 0
  • armorMilitaryBoots, armorSwatHelmet (never hit): floor((20 - 0)/10) = floor(2.0) = 2
• armorMilitaryHelmet gets no nudge
• armorMilitaryBoots and armorSwatHelmet get 2x the nudge factor of 0.05 added to their "prob" attribute
• Probabilities:
  • Total denominator is 1.0 (helmet) + 1.10 (boots) + 0.50 (swat helmet) = 2.60
  • armorMilitaryHelmet = 1.0/2.60 = 38.5%
  • armorMilitaryBoots = 1.1/2.60 = 42.3%
  • armorSwatHelmet = 0.50/2.60 = 19.2%**

Assume the player gets armorMilitaryBoots in this loot.

• lastHitDay for armorMilitaryBoots is set to 20

itemName	firstCheckedDay	lastHitDay
armorMilitaryHelmet	11	15
armorMilitaryBoots	11	20
armorSwatHelmet	11	0

 

6. Day 25 (last one I promise): another looting

• armorMilitaryHelmet (last hit day 15) would get 1x nudge, armorMilitaryBoots (hit on day 20) would get 0 nudge, armorSwatHelmet would get 2x nudge.

* the firstCheckedDay column isn't really required, but it makes for easier demonstration. Really, the system could put -1 into lastHitDay to indicate "never been checked" status, then change it to 0 the very first time the lootgroup has an RNG chance.

** one weakness of adding a fixed "prob" adjustment is that rare items will get a proportionally greater "nudge" than default items. This would be fixed by adding a percentage of the items default "prob" value instead of a fixed amount. For example if it is set to add a 5% nudge, then a 1.0 (default) prob item would get 0.05 added, whereas a 0.4 prob item (like the swat helmet) would get 0.02 (5% of 0.4).

 

 

Yes, I do have the day off work, how did you know?