by critor » 13 Dec 2020, 19:07
@Pavel Appeler la fonction
rnd() de
web.py ne gêne pas a priori.
Ce qui dérange, c'est quand tu l'appelles en boucle un certain nombre de fois afin de garantir que le prochain tirage te sera favorable.
Parce qu'à partir de ce moment-là tu connais
(ou presque) la graine aléatoire, et pourrais parfaitement remonter en arrière et déterminer toute la configuration du volcan sans avoir à le parcourir.
Voici le nouveau
web.py d'évaluation, protégé contre la manipulation de la chance, et utilisé à compter de maintenant :
- Code: Select all
#cas
from math import pi, cos, sin, floor
import sys
#---
# RNG
rnd_seed = int(sys.argv[1]) if len(sys.argv) > 1 else 0xc0ffee
web_dim = int(sys.argv[2]) if len(sys.argv) > 2 else 36
web_density = float(sys.argv[3]) if len(sys.argv) > 3 else .05
pits_density = float(sys.argv[4]) if len(sys.argv) > 4 else .1
bats_density = float(sys.argv[5]) if len(sys.argv) > 5 else .15
def rnd():
global rnd_seed
rnd_max = 0x7fff
rnd_seed = (rnd_seed * 214013 + 2531011) % 4294967296
return ((rnd_seed // (2*rnd_max + 1)) & rnd_max)
def random():
return rnd() / 0x7fff
def randint(a,b):
return rnd() % (b-a+1) + a
def choice(l):
return l[randint(0, len(l)-1)]
#---
screen_h = 240
m_p, m_l, m_k, m_b, m_d, m_a, m_m = 1, 4, 16, 64, 256, 1024, 4096
def insertinto(l1, l2):
for v in l1:
if v not in l2:
l2.append(v)
return l2
def removefrom(l1, l2):
for v in l1:
try:
l2.remove(v)
except:
pass
return l2
def connectPlatforms(s1, s2):
global web
web[s1][s2], web[s2][s1] = 1, 1
def get_reachable_platforms_from_platforms(l, safe):
lv = []
for s in l:
for i in range(dimweb):
if web[s][i]:
if i not in lv and (not(safe) or not (platforms[i] & m_p)):
lv.append(i)
return lv
def cango(s1, s2, safe):
lvo1, lvi1, lvo2, lvi2, t_inter, k = [], [s1], [], [s2], 0, 0
while not (t_inter) and len(lvi1) and len(lvi2):
lvo1, lvo2 = insertinto(lvo1, lvi1), insertinto(lvo2, lvi2)
for v in lvo1:
if v in lvo2:
return k
lvi1, lvi2 = get_reachable_platforms_from_platforms(lvo1, safe), get_reachable_platforms_from_platforms(lvo2, safe)
lvi1, lvi2 = removefrom(lvo1, lvi1), removefrom(lvo2, lvi2)
k += 1
return 0
def my_bitor(a, b):
return ~(~a & ~b)
def init_web(d, p_p, p_b):
global web, platforms, mwspr, mhspr, zoom, mwc, screen_w, screen_h
yweb, l0 = screen_h / 2, list(range(dimweb))
l0.remove(0)
web, platforms, conn, dconn, i_k = [], [0 for k in range(dimweb)], [0], list(range(1, dimweb)), choice(l0)
for j in range(dimweb):
web.append([0 for k in range(dimweb)])
while len(dconn):
s = dconn[randint(0, len(dconn) - 1)]
connectPlatforms(conn[randint(0, len(conn) - 1)], s)
dconn.remove(s)
conn.append(s)
for j in range(dimweb-1):
for i in range(j + 1, dimweb):
if floor(d + random()):
connectPlatforms(i, j)
i_d = choice(l0)
platforms[i_d] = my_bitor(platforms[i_d], m_d)
l1 = list(l0)
for v in get_reachable_platforms_from_platforms([0], 0):
l1.remove(v)
if not(len(l1)):
l1 = l0
l2 = list(l1)
for v in get_reachable_platforms_from_platforms(get_reachable_platforms_from_platforms([0], 0), 0):
try:
l2.remove(v)
except:
pass
if not(len(l2)):
l2 = l1
i_l = choice(l2)
platforms[i_l] = my_bitor(platforms[i_l], m_l)
platforms[i_k] = my_bitor(platforms[i_k], m_k)
for i in l1:
if i != i_k and i != i_d and floor(p_p*dimweb/len(l1) + random()):
platforms[i] = my_bitor(platforms[i], m_p)
if not cango(0, i_k, 1) or not cango(0, i_d, 1):
platforms[i] = platforms[i] & ~m_p
if floor(p_b*dimweb/len(l1) + random()):
platforms[i] = my_bitor(platforms[i], m_b)
def parcourir_selon(ia):
global dimweb, platforms, web_dim, web_density, pits_density, bats_density, rnd_seed
dimweb = web_dim
maxcoups = dimweb**2 * 2
init_web(web_density, pits_density, bats_density)
s0, s1, s2, s3, s4, s5, s6, s7 = 0, 0, m_a, 0, 1, -1, 0, 0
pfs0, pfs5 = platforms[s0], 0
ex_infos = ""
while s4 > 0 and (not (s2 & (2 * m_k)) or not (pfs0 & m_d)):
if s5 < 0:
s5 = 0
else:
try:
rnd_seed_backup = rnd_seed
k, k2 = ia(s0, voisines, dimweb, s1, s2)
rnd_seed = rnd_seed_backup
if pfs5 & (2 * m_b):
while s0 == s5:
s0 = randint(0, dimweb - 1)
pfs0, pfs5 = my_bitor(platforms[s0], m_b), pfs5 & ~(3 * m_b) & ~m_m
else:
if k2:
if s2 & m_a:
v = platforms[k]
if v & m_l:
v, s2 = v & ~m_l, my_bitor(s2, 2 * m_l)
platforms[k] = my_bitor(v, 2 * m_l)
s2 = s2 & ~m_a
s2 = my_bitor(s2, 2 * m_a)
else:
if k in voisines:
s0 = k
if pfs5 & m_b:
pfs5 = my_bitor(pfs5, 2 * m_b)
pfs0, pfs5 = platforms[s0], pfs5 & ~m_m
s3 += 1
if s3 >= maxcoups:
s4 = 0
if pfs0 & m_k:
pfs0 = pfs0 & ~m_k
s2 = my_bitor(s2, 2 * m_k)
if pfs0 & my_bitor(m_p, m_l):
s4 = 0
pfs0 = my_bitor(pfs0, 2 * m_m)
platforms[s5] = pfs5
except Exception as t_excpt:
s4 = -1
ex_infos = str(t_excpt)
print(t_excpt)
pfs0 = my_bitor(pfs0, m_m)
s1, voisines = pfs0, get_reachable_platforms_from_platforms([s0], 0)
platforms[s0] = pfs0
for v in voisines:
t = my_bitor(m_p, m_l)
t = platforms[v] & my_bitor(t, m_k)
s1 = my_bitor(s1, t)
for v in get_reachable_platforms_from_platforms(voisines, 0):
t = platforms[v] & m_l
s1 = my_bitor(s1, t)
s, sold = my_bitor(s1, s2), my_bitor(s6, s7)
s5, s6, s7, pfs5 = s0, s1, s2, pfs0
r = s4 > 0 and s3 < maxcoups
if r:
print("win " + str(s3), file=sys.stderr)
else:
if pfs0 & m_l:
print("leviathan", file=sys.stderr)
elif pfs0 & m_p:
print("well", file=sys.stderr)
elif s3 >= maxcoups:
print("lowpower", file=sys.stderr)
elif s4 < 0:
print("exception " + ex_infos, file=sys.stderr)
return r, s2, s3
Les changements sont dans
parcourir_selon(ia), 2 lignes ajoutées autour de l'appel à
ia().
D'ailleurs l'IA invalide qui n'a pas terminé son cycle de 100000 a déjà commencé à chuter et n'est plus 1ère, suite à la mise en place de cette nouvelle version.
On lui relancera le cycle au complet quand elle aura terminé de toutes façons.
Et peut-être pour tout ou partie des autres IA aussi, on verra.
...
)
