要提升微信看一看推荐混排的长期收益？试试深度强化学习

import gym

import random

import numpy

N_BINS = [5, 5, 5, 5]

LEARNING_RATE=0.05

DISCOUNT_FACTOR=0.9

EPS = 0.3

MIN_VALUES = [-0.5,-2.0,-0.5,-3.0]

MAX_VALUES = [0.5,2.0,0.5,3.0]

BINS = [numpy.linspace(MIN_VALUES[i], MAX_VALUES[i], N_BINS[i]) for i in xrange(4)]

def discretize(obs):

    return tuple([int(numpy.digitize(obs[i], BINS[i])) for i in xrange(4)])

qv = {}

env = gym.make('CartPole-v0')

print(env.action_space)

print(env.observation_space)

an = env.action_space.n

def get(s, a):

    global qv

    if (s, a) not in qv:

        return 0

    return qv[(s, a)]

def update(s, a, s1, r):

    global qv

    nows = get(s, a)

    m0 = get(s1, 0)

    m1 = get(s1, 1)

    if m0 < m1:

        m0 = m1

    qv[(s, a)] = nows + LEARNING_RATE * (r + DISCOUNT_FACTOR * m0 - nows)

for i in range(500000):

    obs = env.reset()

    if i % 1000 == 0:

        print i

    for _ in range(5000):

        s = discretize(obs)

        s_0 = get(s, 0)

        nowa = 0

        s_1 = get(s, 1)

        if s_1 > s_0:

            nowa = 1

        if random.random() <= EPS:

            nowa = 1 - nowa

        obs, reward, done, info = env.step(nowa)

        s1 = discretize(obs)

        if done:

            reward = -10

        update(s, nowa, s1, reward)

        if done:

            break

for i_episode in range(1):

    obs = env.reset()

    for t in range(5000):

        env.render()

        s = discretize(obs)

        maxs = get(s, 0)

        maxa = 0

        nows = get(s, 1)

        if nows > maxs:

            maxa = 1

        obs, reward, done, info = env.step(maxa)

        if done:

            print("Episode finished after {} timesteps".format(t+1))

            break