[PYTHON] Reinforcement learning 18 Colaboratory + Acrobat + ChainerRL

It is assumed that reinforcement learning 17 has been completed. I've summarized it first to make it easier to change the parameters. It takes about 30 minutes to study 200 times. At this level, it was slow when using the GPU. CPU is recommended. If you study for a long time, you will get caught up in the 12-hour rule. It seems good to divide it into small pieces. If you want to subdivide, after learning 100 times, use it as a backup Like agent.save (save_point_num). If you want to start from the middle It should be like agent.load (last_save_point).

Google Drive mount

import google.colab.drive
google.colab.drive.mount('gdrive')
!ln -s gdrive/My\ Drive mydrive

Installation

!apt-get install -y xvfb python-opengl ffmpeg > /dev/null 2>&1
!pip install pyvirtualdisplay > /dev/null 2>&1
!pip -q install JSAnimation
!pip -q install chainerrl

Parameter initialization

gamename='Acrobot-v1'
# Set the discount factor that discounts future rewards.
gamma = 0.99
# Use epsilon-greedy for exploration
myepsilon=0.03
myDir='mydrive/OpenAI/Acrobot/'
mySteps=200000 # Train the agent for 2000 steps
my_eval_n_episodes=1 # 10 episodes are sampled for each evaluation
my_eval_max_episode_len=200  # Maximum length of each episodes
my_eval_interval=1000   # Evaluate the agent after every 1000 steps
myOutDir=myDir+'result'      # Save everything to 'result' directory
myAgentDir=myDir+'agent'      # Save Agent to 'agent' directory
myAnimName=myDir+'movie_acrobot.mp4'
myScoreName=myDir+"result/scores.txt"

Program

import

import chainer
import chainer.functions as F
import chainer.links as L
import chainerrl
import gym
import numpy as np

env initialize

env = gym.make(gamename)
print('observation space:', env.observation_space)
print('action space:', env.action_space)

obs = env.reset()
print('initial observation:', obs)
action = env.action_space.sample()
obs, r, done, info = env.step(action)
print('next observation:', obs)
print('reward:', r)
print('done:', done)
print('info:', info)

Deep Q Network setting

obs_size = env.observation_space.shape[0]
n_actions = env.action_space.n
q_func = chainerrl.q_functions.FCStateQFunctionWithDiscreteAction(
    obs_size, n_actions,
    n_hidden_layers=2, n_hidden_channels=50)

Use Adam to optimize q_func. eps=1e-2 is for stability.

optimizer = chainer.optimizers.Adam(eps=1e-2)
optimizer.setup(q_func)

Agent Setting DQN uses Experience Replay.

Specify a replay buffer and its capacity.

Since observations from CartPole-v0 is numpy.float64 while

Chainer only accepts numpy.float32 by default, specify a converter as a feature extractor function phi.

explorer = chainerrl.explorers.ConstantEpsilonGreedy(
    epsilon=myepsilon, random_action_func=env.action_space.sample)
replay_buffer = chainerrl.replay_buffer.ReplayBuffer(capacity=10 ** 6)
phi = lambda x: x.astype(np.float32, copy=False)
agent = chainerrl.agents.DoubleDQN(
    q_func, optimizer, replay_buffer, gamma, explorer,
    replay_start_size=500, update_interval=1,
    target_update_interval=100, phi=phi)

Train

Set up the logger to print info messages for understandability.

import logging
import sys
logging.basicConfig(level=logging.INFO, stream=sys.stdout, format='')
chainerrl.experiments.train_agent_with_evaluation(
    agent, env,steps=mySteps,eval_n_steps=None,eval_n_episodes=my_eval_n_episodes,eval_max_episode_len=my_eval_max_episode_len,
    eval_interval=my_eval_interval,outdir=myOutDir)
agent.save(myAgentDir)

Data Table

import pandas as pd
import glob
import os
score_files = glob.glob(myScoreName)
score_files.sort(key=os.path.getmtime)
score_file = score_files[-1]
df = pd.read_csv(score_file, delimiter='\t' )
df

figure Average_Q

df.plot(x='steps',y='average_q')

Test

import2

from pyvirtualdisplay import Display
display = Display(visible=0, size=(1024, 768))
display.start()


from JSAnimation.IPython_display import display_animation
from matplotlib import animation
import matplotlib.pyplot as plt
%matplotlib inline

Test Program

frames = []
env = gym.make(gamename)
envw = gym.wrappers.Monitor(env, myOutDir, force=True)

for i in range(3):
    obs = envw.reset()
    done = False
    R = 0
    t = 0
    while not done and t < 200:
        frames.append(envw.render(mode = 'rgb_array'))
        action = agent.act(obs)
        obs, r, done, _ = envw.step(action)
        R += r
        t += 1
    print('test episode:', i, 'R:', R)
    agent.stop_episode()
#envw.render()
envw.close()

from IPython.display import HTML
plt.figure(figsize=(frames[0].shape[1]/72.0, frames[0].shape[0]/72.0),dpi=72)
patch = plt.imshow(frames[0])
plt.axis('off') 
def animate(i):
    patch.set_data(frames[i])
anim = animation.FuncAnimation(plt.gcf(), animate, frames=len(frames),interval=50)
anim.save(myAnimName)
HTML(anim.to_jshtml())