mdqn.py

#!/usr/bin/env python
import signal
import torch
import os
from environment import Environment
import datageneration
import train 
import time
import sys
import subprocess
from subprocess import Popen
from os.path import abspath, dirname, join
import config as cfg


def getValue(filename):
	line = subprocess.check_output(['tail', '-1', filename])
	return int(line.decode('utf-8').replace('\n', ''))

def setValue(filename,value):
	f = open(filename, "w")
	f.write(str(value))
	f.close()


def openSim(process):
	process.terminate()
	time.sleep(5)
	process = Popen(command)
	time.sleep(5)
	return process

def killSim(process):
	process.terminate()		
	time.sleep(10)

def signalHandler(sig, frame):
    process.terminate()
    sys.exit(0)


t_episodes=cfg.t_episodes

file_phase = 'files/phase.txt'

episode=int(torch.load('files/episode.dat'))

command = './simDRLSR.x86_64'
execute_simulator = False
if(len(sys.argv)>1):
	execute_simulator = True
	directory = str(sys.argv[1])
	command = abspath(join(directory,command))
		#command = directory+command


phase = getValue(file_phase)
process = Popen('false') # something long running
signal.signal(signal.SIGINT, signalHandler)

print(episode,t_episodes)

for i in range(episode,t_episodes+1):
	phase = getValue(file_phase)

	if(phase == 0):
		print("Episode: ",i," collection data.")

		if execute_simulator: process = openSim(process)
		recent_rewards=torch.load('recent_rewards.dat')
		reward_history=torch.load('files/reward_history.dat')
		print(len(recent_rewards))
		
		env=Environment(epi=i)

		env.send_data_to_pepper("start")
		time.sleep(1)
		env.close_connection()
		time.sleep(1)
		#Execute data generation phase script
		datageneration.main()

		setValue(file_phase,1)

		env=Environment(epi=i)
		env.send_data_to_pepper("stop")



	phase = getValue(file_phase)
	if(phase == 1):
		print("Episode: ",i," training model.")
		
		print("Sending signal to kill simulator")
		if execute_simulator: killSim(process)
		setValue('flag_simulator.txt',9)
		time.sleep(1)		
		#Execute train phase script
		train.main()

		setValue(file_phase,0)

	if execute_simulator: killSim(process)
print("Model trained...")