Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[NRDL]

PTW

[wierd]

I am looking at the therapist numbers here, and am unsure how to interpret them.

I see two sets of numbers for each physical attribute. How should I be trying to read these?  As presented, they look like a ratio presented as a fraction, but each dwarf has a different high/low?  When treated as a decimal ratio of 1, they make more sense than in the fashion presented, is that the way I should be reading it?  Closer to 1 means higher physical attribute, and closer to 0 means lower physical attribute, with a .5 being the median of mediocrity?

[Jurojin00]

Current observations are indicative that dwarfs do inherit traits correct?

I have a mind to establish a research facility to carry out the tests as well, I'm between forts and have a lot of free time at the moment.

I do not believe dwarfs inherit traits. I created a fort to do some research into attribute inheritance which is not completed yet, but since I have the data already I decided to look into the traits to check whether they are inheritable or not.

Setup:

Spoiler (click to show/hide)

DF 34.11 with the following modifications:

• [NO_EAT]
• [NO_DRINK]
• [NOEMOTION]
• [MAXAGE:10:10]
• [LITTERSIZE:10000:10000]
• removed [MULTIPLE_LITTER_RARE]

Dwarfs were allowed to choose their own partner. Data comes from a single litter (2 Parents, 10000 Children).
Data exported using Dfhacks "dwarfexport". A few regular expressions to allow me to import the xml into Excel which was used for analysis.

Average Traits:

Spoiler (click to show/hide)

As you can see, the traits are just 50 on average with negligible variation. There does not seem to be any correlation between the parents and children.

Trait Histogram:

Spoiler (click to show/hide)

(Explanation: The X axis shows the upper limit of the group so the bar at 2 on the X axis includes all the values >= -2 and < 2)
The distribution of all the traits is pretty much identical.

Since there is alot more going on with attributes, I still have to gather some additional data before showing something. I am happy to share the RAW data if anyone is interested.

PS.: Atom-Smashing 10000 dwarven babys will cause DF to crash every time. So if you want to gather more then one sample in a given fort you should savescum after exporting and smash the mother right before birth. As you might imagine, FPS are not great with so many dwarfs.

[wierd]

Interesting approach--  increase litter size, shorten lifespan.

How many generations did you allow to run?

Still, I too am not seeing much of a relationship between parents and offspring that would account for attributes they are born with, but with a much smaller sample size.

[Jurojin00]

It was just one generation. I just want to see what the game does without major interference on my part. I just shortened the lifespan so I donīt have to worry about age when selecting breeding pairs. Unlike with traits, the attributes appear to depend on the parents more but I want to do a few more litters so I can be reasonably sure of my conclusions before posting anything.

EDIT: Multiple generations are not possible with this setup as you can not get rid of the excess babies without major effort or without crashing the game. Also, I donīt think the fort would survive long when it is haunted by 10000 baby ghosts.

[wierd]

Try using LAA flags on DF.  It can then allocate 3gb of address space on a 64bit platform, instead of 2gb.  May help with the absurd baby explosion rate.

Part of the experiment is to track attribute inheritence over several generations, using the culling rate as a metric. if I set the litter size higher for awhile, it would allow me to get more offspring per popping, and thus have larger generational populations after the aggressive culling.  I am currently worried that the culling rate will be too high to sustain the fortress, as it is currently above 50% of all offspring produced!

[Jurojin00]

I got some discouraging results regarding attribute inheritance.
Setup is same as before.

Setup:

Spoiler (click to show/hide)

DF 34.11 with the following modifications:

• [NO_EAT]
• [NO_DRINK]
• [NOEMOTION]
• [MAXAGE:10:10]
• [LITTERSIZE:10000:10000]
• removed [MULTIPLE_LITTER_RARE]

Dwarfs were allowed to choose their own partner. Data comes from a single litter (2 Parents, 10000 Children).
Data exported using Dfhacks "dwarfexport". A few regular expressions to allow me to import the xml into Excel which was used for analysis.

Attribute means:

Spoiler (click to show/hide)

There first appeared to be an obvious correlation between the parents and their children, but that this was not due to them inheriting their parents attributes, but because all dwarfs attributes are chosen from the same distributions. As you can see the mean attribute values of the children are virtually identical, suggesting that there is no inheritance of attributes. I generated a different world for Parents/Litter 4 to make sure, that the attribute generation does not depend on the world.

[Snaake]

Another reason to use [CHILD:0] instead of 1 would be that in another recent thread, it was pointed out that abilities rust as well, not just skills (someone was wondering why their children were becoming dumber with age). 1 year isn't really a long time for ability rust I guess, but it's still a systemic error that could prevent or slow down the generational upwards trend in abilities that I guess you're trying to achieve/prove/test.

And about litter size, even just setting your dwarves to always have twins (that pop out fully grown!) should almost double your chances of good kids. Or quadruplets. Or if you still have a baseline save with the groups established but without actual breeding done, a larger litter size accelerated trial   AFAIK you still need to balance culling with the parents' mental health (even if the only bad thought is the death of the child, no rotting or anything)?

I got some discouraging results regarding attribute inheritance.
...
Attribute means:

Spoiler (click to show/hide)

There first appeared to be an obvious correlation between the parents and their children, but that this was not due to them inheriting their parents attributes, but because all dwarfs attributes are chosen from the same distributions. As you can see the mean attribute values of the children are virtually identical, suggesting that there is no inheritance of attributes. I generated a different world for Parents/Litter 4 to make sure, that the attribute generation does not depend on the world.

Wow, discouraging indeed. I would suggest checking whether your initial 7 also conform to the same dwarven racial attribute distribution (using the standard RPG definition of "race", not the IRL one; i.e. species, not breed)? I think there's a command in dfhack to embark with more than 7 dwarves that sounds like it would be useful for testing this (I really haven't used dfhack basically at all, so don't actually know). Immigrants that are historical persons could very possibly have their stats skewed according to what skills they're generated with.

(I was also going to suggest testing a version where you play matchmaker for your dwarves, trying for a couple with high strength etc., but I guess that would still give the same results, since your attribute for each stat are identical for all the litters)

[Jurojin00]

I would suggest checking whether your initial 7 also conform to the same dwarven racial attribute distribution (using the standard RPG definition of "race", not the IRL one; i.e. species, not breed)? I think there's a command in dfhack to embark with more than 7 dwarves that sounds like it would be useful for testing this (I really haven't used dfhack basically at all, so don't actually know). Immigrants that are historical persons could very possibly have their stats skewed according to what skills they're generated with.

All of the previous parents were migrants so I just smashed the other migrant couples and waited for the starting 7 to produce children. The result is the same unfortunately.

[Sutremaine]

I see two sets of numbers for each physical attribute. How should I be trying to read these?  As presented, they look like a ratio presented as a fraction, but each dwarf has a different high/low?

Not 100% sure where you're looking, but each attribute has two numbers. The first is the current stat, the second is the max stat. A dwarf's maximum in a stat is either its starting stat + the racial average or 2x the starting stat, whichever is higher. Urist McSlowpoke will have something like 503/1400 (503 current, 1400 max, has gained 3 Agility), and Urist McSpeedy will have something like 1268/2400 (1268 current, 2400 max, has gained 68 Agility). All dwarves will have respectable max stats for Focus and Spatial Sense due to the racial average of 1500, and Agility will tend to be the lowest due to the average of 900.

[wierd]

gotcha.  that just seems a strange way to store and track the data, when a float between 0 and 1, with weighted algos would make more sense, at least to me anyway.  That way calculations made using them would be floating point ops, and not need to tie up ALUs on the general purpose side of the CPU. (EG, could use SSE* and pals) It would also use simpler program code.

but meh.

I am agreeing with my recent conspirator however; I am not seeing any real correlations jumping out at me.

I will run the experiment to completion anyway, however. Even if the only reason why there is a punctured equilibrium in the fortress is because I have been brutally butchering babies, the result of having a large and statistically improbable sampling of superior dwarves would still make for a good succession game on this world later.

If anything, the fact that this is basically a lesson in cat herding just makes the process even MORE infamous.

If nothing else, we can still use the test apparatus to answer other reproductive questions, like at what degree of familial separation marriage becomes allowable again, and how skin and hair color are inherited, weather fattness is heritable, etc.

[krenshala]

I am looking at the therapist numbers here, and am unsure how to interpret them.

I see two sets of numbers for each physical attribute. How should I be trying to read these?  As presented, they look like a ratio presented as a fraction, but each dwarf has a different high/low?  When treated as a decimal ratio of 1, they make more sense than in the fashion presented, is that the way I should be reading it?  Closer to 1 means higher physical attribute, and closer to 0 means lower physical attribute, with a .5 being the median of mediocrity?

I believe that is the "current/max for that dwarf" attribute value.  So, 2584/4000 would mean he has an attribute score of 2584 of a possible 4000 (for him).

* And, I see someone else already answer this better.

[Sutremaine]

Yeah, I was just using Agility as an example of a particular stat. A dwarf has lots of stats (or attributes; I seem to be using either word interchangeably), and each one has its own current and maximum number.

[wierd]

I just meant that making each dwarf have variant max values means you perform a lot of checks on the data before you can even process it that are logic checks, not math checks.  It would make more sense to abstract that as a mathematical threshold by assigning variables: the first as a float value between 0 and 1, and the second an 8 bit flag variable to store some information about the standard deviation that dwarf belongs in (which would dynamically set the max value using pure math), etc.

That way the main ALUs in the CPU would be freed up from checking things like EvaluateDwarf(CurrentStat,MaxStat,MentalImplicationMatrix,CurrentActivity) which would need the general logic abilities of the CPU.  It could instead be done as a pure math operation with SSE or the FPU instead, for a rather profound increase in execution, given the absurd number of times it gets fired.

It would also use significantly less memory.

but, Meh.

[Snaake]

...I will run the experiment to completion anyway, however. Even if the only reason why there is a punctured equilibrium in the fortress is because I have been brutally butchering babies, the result of having a large and statistically improbable sampling of superior dwarves would still make for a good succession game on this world later.
...

Hm. A tower of evil, inhabited by zombies, a large part of them children, surrounded by a massive fortress still inhabited by their physically and mentally superior parents, siblings, nephews and nieces (and so on for however many generations you go on). Were the live dwarves twisted by the evil in the tower, or are they so evil that they have enslaved the necromancers for their own purposes?

Very dwarfy. Would be a cool place to go as an adventurer, once fortress retiring is in?