Very simple random search

CartPole-v0.py

# from http://kvfrans.com/simple-algoritms-for-solving-cartpole/

import gym
from gym import wrappers
import numpy as np

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

def run_episode(env, parameters):  
    observation = env.reset()
    totalreward = 0
    for _ in xrange(2000):
        action = 0 if np.matmul(parameters,observation) < 0 else 1
        observation, reward, done, info = env.step(action)
        totalreward += reward
        if done:
            break
    return totalreward
  
parameters = np.random.rand(4) * 2 - 1  
bestparams = None  
bestreward = 0  
episodes_per_update = 5
for _ in xrange(10000):  
    parameters = np.random.rand(4) * 2 - 1
    reward = 0
    for _ in xrange(episodes_per_update):  
        run = run_episode(env,parameters)
        reward += run / episodes_per_update
    if reward > bestreward:
        bestreward = reward
        bestparams = parameters

        if reward >= 2000:
            break
print bestreward

env = gym.make('CartPole-v0')
env = wrappers.Monitor(env, '/tmp/cartpole-experiment-1', force=True)
for i_episode in range(100):
    print i_episode
    observation = env.reset()
    for _ in xrange(1000):
        #env.render()
        action = 0 if np.matmul(parameters, observation) < 0 else 1
        observation, reward, done, info = env.step(action)
        if done:
            break
env.close()