general.py

import gym
env = gym.make('CartPole-v0')
obs=env.reset()
maxs=[obs[i] for i in range(len(obs))]
mins=[obs[i] for i in range(len(obs))]

def strat(style):
    global maxs, mins
    ss=lambda x: 0
    if style=="1":
        ss=lambda x: 1
    if style=="a":
        ss=lambda x: x%2
    if style=="r":
        ss=lambda x: env.action_space.sample()
    obs=env.reset()
    for i in xrange(1000):
        move=ss(i)
        obs,r,d,i=env.step(move)
        if d:
            return
        for i in range(len(obs)):
            if obs[i]>maxs[i]:
                maxs[i] = obs[i]
            if obs[i]<mins[i]:
                mins[i]=obs[i]

for i in xrange(2000):
    strat("0")
for i in xrange(2000):
    strat("a")
for i in xrange(2000):
    strat("1")
for i in xrange(2000):
    strat("r")
print maxs
print mins

def test(wts, maxrounds=300):
    global maxs, mins
    ss=lambda x: sum([wts[i]*x[i]/maxs[i] for i in range(len(x))]) > 0
    rwd=0
    obs=env.reset()
    for i in xrange(maxrounds):
        move=1 if ss(obs) else 0
        obs,r,d,i=env.step(move)
        rwd += r
        if d:
            return rwd
    return rwd

themax=0.0
import random
roughs = [-1 + .2*i for i in range(11)]
random.shuffle(roughs)

for x1 in roughs:
    for x2 in roughs:
        for x3 in roughs:
            for x4 in roughs:
                rwd1 = test([x1, x2, x3, x4])
                rwd2 = test([x1, x2, x3, x4])
                rwd3 = test([x1, x2, x3, x4])
                rwd = (rwd1 + rwd2 + rwd3)
                if rwd > themax:
                    wts = [x1, x2, x3, x4]
                    themax=rwd
                    print rwd, wts

trial = [test(wts, 200) for i in range(100)]
print sum(trial)/len(trial)
print wts

def ai():
    global wts, maxs
    obs=env.reset()
    rwd=0
    for i in range(200):
        move=1 if sum([obs[i]*wts[i]/maxs[i] for i in range(len(obs))]) > 0 else 0
        obs,r,d,i=env.step(move)
        rwd+=r
        if d:
            return rwd
    return rwd