as of oct 26, 2016: i have posted updates on the mailing list, long ago as of sep 19, 2016: post racketcon '16: i'd try to wrap this up better then post on mailing list
racket -tm pd-population.rkt
how to extract automata: ... (cont)
so vincent emailed me
and i started again the simulation, under his advision
some notable features are:
- the data structure of automata is represented in hash table
- immutable mutation
- in interaction, delta is the continual probability
- redo the population (previous: double vectored population)
- mutation: add and delete states
- try to minimize matha in the work flow (export automata code, solve games w matrix form, plot replicator dynamics etc)
population of finite state machines playing the repeated prisoner's dilemma. a sample code.
- only start to log at 22 august 2016
at this point, the state of the code is as follows:
automata.rkt
- basic automaton functions
- classic automata
- interaction (with delta as continual probability)
- immutable mutation (hash)
- holy god, at 31 aug i fix the bug #2
error:
hash-ref: no 'D key
blame: interact...
i have to output population at each cycle in the `evolve
then print "I-AM-HERE!!!!!" everytime it throws exception
then print output of `interact, with handlers, so if it throws exception, it still returns the automaton pair
i discover that the cause is that somehow the states in the troublesome automaton gets an extra pair of parentheses (OMG)
i guess it is due to `mutate
then print output of `mutate with its ID
then i discover that it is the `detach state that adds the extra pair
because hash->list and list->hash i have to add `flatten
- 7 sep: work flow to develop the benchmark automata in the personality test
ressurect at
then find a way to test their personality
then collect the useful benchmarks into a table
that would be materials for selecting an official benchmark list
| ID | To do | Done? | Date | by | Note |
|---|---|---|---|---|---|
| 1 | redo the population | x | 0309 | chi | works ..ok (shows the cooperative periods) |
| 2 | evolve doesnt work: hash-ref value not found: interaction | x | 3108 | blame: mutation, because the exception is thrown at arbitrary settings | |
| 3 | export to graphviz dot instead of matha code | x | 24 | with-output-to-dot | |
| 4 | reset: both current and payoff | x | 23 | ||
| 5 | visualise the mutation process on TV | ugh later.. | |||
| 6 | why it doesnt show cooperative period | x | 0109 | try to test the rd first, wo mutation: fail. blame: `scan right, but forget to reset payoff before scan | |
| 7 | replicator dynamics test looks squiggly ugly | blame: continual probability method. change to discount method, still ugly | |||
| 8 | from #6, if rd test passed, why it doesnt show cooperative period? | x | 2 | STILL DOESNT SHOW COOPERATIVE PERIOD??? WTF. finally it shows, but with very high mutation rate | |
| 9 | the characteristic test | ||||
| 10 | see how to export w printf in the link vincent sent | o | with printf, you dont have to string-append. just print, and with-output-to-file will collect them into output file. however, i print different parts with `for and plug them together. so.. task aborted. | ||
| 11 | from #5: show on tv the ranking then print them out over cycles | pr'bly a combination of plot/dc or draw w graphviz, #:exist 'append | |||
| 12 | in each cycle, scan -> export dot codes together with statistics | x | 5 | produce multiple graphs in .ps then pdf (dot -Tps2 ... | |
| 13 | bc i have to export automata into txt then recover -> it's better (for the later reading) to flatten them first | x | 5 | ||
| 14 | export: odd line - cycle, even line - all the rankings in one line, if null leave an empty line | x | 6 | easier for import | |
| 15 | now import & recover automata | x | 7 | ||
| 16 | ressurect-at | x | 8 | ||
| test personality at cycle 2000, sample | |||||
| get matha out of the loop | |||||
| revise code, in light of these new discoveries haha: list-update, list-set, [j (in-naturals), [e l] ~ [e (in-list l)]] | |||||
| write br function, solve NE | x | 9 | |||
| run semi extensive simulations to get patterns, in order to develop benchmark set for the personality test | |||||
| export graphviz to pdf: big automata -> dont fit into A4 page of pdf, but postscript file is ok | |||||
| generate file names reflecting different deltas | |||||
| run across deltas | x | 9 | |||
| you dont have time for developing the personality test. just see what happens in the special periods | :((( | ||||
| generate more data to see the patterns as delta runs along the spectrum (middle high deltas: more interesting, delta close to 1: mostly tit for tat) | |||||