2000-08 Intervening runs


Runs without stats for some reason?

For some reason, I have no statistic files for the runs I did between 2000-07 and 2000-10. I think perhaps this was where I had made the spectacular mistake of deleting data for exactly those runs that I had used for the paper, but it seems a bit early. The full list of runs that I believe are from this period are below.

I note that from about 1428 on I was, for some reason, trying to replicate the lisp results. The reason this puzzles me is that I was already getting the right kind of suicide rate, but I recall going back to replicating results because of a lack of something or other that was in the lisp results. I did have a few bugs, but they were nothing that would substantially affect the outcome.

By 1567, I've returned to my c++ code. With 1617, I've made a whole swag of changes

1679 - Testing visualizations by genetic relatedness to randomly selected agent

1633 - Testing visualizations by age

1632 - Testing visualizations by age (had walking errors)

1619 - Board Size: 80x80, full cell 1, mate neighbourhood 5x5

1618 - Initial Agents: 4000, food offset: 140 (birth rate roughly 1/3 death rate initially, dropping to 1/5 or less later)

1617 - Many changes - negative probabilities, big board 100x100, mutations allowed in ActionProbabilityVectors, Limited full cell size to 2, decreased health effect of walking, re-enabled resting, small negative health effect for failing action, age limit of food, 8 cycles - still 60 mag 60 offset food. (they all died out). Need to increase birth rate about five-fold to match death rate.

1567 - cyclical, no suicide

<<switch to c++>>

1566 - cyclical, std

1565 - constant, std

1564 - constant, crash (pos. cause: 1x1 neighborhood)

1563 - test run

<<switch to lisp>>

1562 - constant, 3x3 neighborhood

1561 - constant, standard (cmp 1560)

1560 - cyclical, standard (cmp 1561)

1559 - constant, standard (compare 1558)

1558 - cyclical, standard 1x1 mate neighborhood (compare 1559)

1557 - constant, 50 agents max

1556 - constant, 40 agents max

1555 - constant, 20 agents max - dies out

1554 - constant, 50 agents max - dies out

1553 - constant, 900 agents max*

1552 - constant, 800 agents max

1551 - constant, 700 agents max

1550 - constant, 600 agents max

1549 - constant, 500 agents max

1548 - constant, 400 agents max

1547 - constant, 300 agents max

1546 - constant, 200 agents max

1545 - constant, 100 agents max

1544 - constant, 50 agents max*

1543 - cyclical, constant food investigation

1558 & 1559 - less promising. constant food rises to the same level that cyclical food rises too. It is interesting to note that constant food in a 3x3 neighbourhood looks as if it may not. Will need to run sims to check this. My next runs, however, are longer version runs of 155[8/9].

1543 & 1538 - comparison of constant and cyclical runs looks promising, but need to unify their parameters, and change only constant/cyclical food distribution

1540 - cyclical, ??

1538 - constant, long run (10000 cycles)

1533 - 1542 - ??

1532 - constant, with still-born probability

1531 - DBG

1514 thru 1528 - constant - forked from 2000-07-30, using max agents 100 thru 1400

thru 1513 - DBG, errors in sim running

14.. - constant, food/cycle 60, max agents 1400, full cell 10000

1481 - constant, food/cycle 60, max agents 100, full cell 10000

1480 - cyclic, replication test

1479 - DBG

1478 - constant, age limit 45

1477 - constant, food health 4000 (die out)

1476 - constant, age limit 40 (die out)

1475 - constant, food/cycle 5, food health 2000, age limit 60 (die out)

1474 - constant, food/cycle 3, food health 2000, age limit 60 (die out)

1473 - constant, food/cycle 5, food health 1000, no max agents

1472 - DBG

1471 - constant, food/cycle 1000, no max agents, food health 30

1470 - constant, max items cell 10000, food/cycle 500, max agents 1000

1469 - constant food, max num of agents when reproducing (1000), try to test when there is no pop. pressures . . .

1456 - cyclic assertion

<<fixed crash>>

1455 - cyclic <crash>

<<modified prob-vectors to remain non-normalized>>

1454 - DBG

1453 - cyclic

<<modified mating to coincide more with the lisp version>>

1452 - cyclic

<<fixed board moving to stop crash - caused an agent to be on the CBEList in shaker, but not on the board>>

1449 thru 1451 - DBG

<<modified walk to turn if it fails, modified turn to behave like lisp (only 1 dir possible), health set to 0 on kill of LE, fixed agent's grow old, even if their turn is taken away>>

1448 - cyclic

<<fix to have APM SD = 0.01 initially, and APM SD for all new borns = 1>>

1447 - cyclic

<<removed optimisation temporarily, since it could cause premature ending of turn-list. Also, made FinaliseStatistics a precycleaction to accord with lisp impl.>>

1441 thru 1446 - DBG

<<modified the way in which turns are had, so that newly added cbe's to not get to execute until the following cycle>>

1440 - cyclic

<<put in more compat. stuff - this time, generate initial pop. with gaussian of APM.sd = 0.01 (but all later agents still have APM.sd of 1)>>

1439 - cyclic, observation extremes for stats collection = 2000

1438 - cyclic, APM.SD = 1

<<modified mutation of default prob-vector to take values from meta-tom, as opposed to agent's apm (to reflect lisp impl)>>

1437 - cyclic, APM.SD = 0.01

1436 - cyclic,

1428 thru 1435 - DBG

<<2000-08-14: focus solely on replication of original results>>

1427 - cyclic, food health to 150

1426 - cyclic, food health to 130

1425 - cyclic, food mag 120 offset 120 (pop stable)

1424 - cyclic, food health to 210 (tripled)

1423 - cyclic, overcrowding (1 item per cell). doubled food health, 60x60 board, age limit lowered to 60

1422 - cyclic - (quit early to test over-crowding idea)

<<added sampling of age and health - should possibly sample more stuff as well>>

1421 - cyclic (to see effect of removal of directional bias, and mutation parameter means = 0)

<<randomized turning and randomized baby's direction on birth>>

1420 - DBG, cyclic, APM.Mean and TVM.Mean = 0

1419 - DBG, cyclic

<<discovered poss. cause of tendencies for agent's to be at top of board - all agent's are born facing north . . . - also, found poss. sol. to dying off pop problem, due to incorrect mutation rate reading>>

1416 thru 1418 - DBG

1415 - DBG, cyclic

<<added count of num children agent has at death - to compare num of children with effective num of children>>

1414 - DBG, cyclic

<<changes to check why population going to 0, if new loci selection in place>>

1413 - DBG, cyclic

<<restored uniform loci selection>>

1412 - DEBUG, cyclic, body sub-lengths mean of 6, sd of 8

<<fix for ClippedGaussian, to stop ignoring the mean and sd>>

1411 - debug

1410 - cyclic

<<fix for sds of generated mutation means and sds to be independent (instead of all having same sds)>>

1409 - cyclic

<<changed fixed biasing for sub-length chosen, not locus chosen>>

1408 - debug

<<changed body length to bias to a particular locus>>

1407 - cyclic

1406 - debug

1405 - debug

<<added count of failed actions, changed avg msd to avg rmsd>>

1404 - cyclic

1403 - cyclic

1402 - cyclic, standard run

1399 - cyclic, max body length 40, global food and agents observations, food health m60 sd3