ap = argparse.ArgumentParser()
ap.add_argument("-n", "--number-servers", required=True, help="number of servers to spawn", type=int)
ap.add_argument("-p", "--ports-start", required=True, help="the start of the range of ports to use", type=int)
ap.add_argument("-t", "--host", default="None", help="the host; default is local host; string either internet domain or IPv4", type=str)
args = vars(ap.parse_args())
number_servers = args["number_servers"]
ports_start = args["ports_start"]
host = args["host"]
if host == 'None':
host = socket.gethostname()
example_environment = resume_env(plot=False, dump_CL=100, dump_debug=1, dump_vtu=50)
use_best_model = True
environments = []
for crrt_simu in range(number_servers):
environments.append(RemoteEnvironmentClient(
example_environment, verbose=0, port=ports_start + crrt_simu, host=host,
timing_print=(crrt_simu == 0) # Only print time info for env_0
))
network = [dict(type='retrieve', tensors = ['obs']), dict(type='dense', size=512), dict(type='dense', size=512)]
agent = Agent.create(
# Agent + Environment
agent='ppo', # Agent specification
ap = argparse.ArgumentParser()
ap.add_argument("-n", "--number-servers", required=True, help="number of servers to spawn", type=int)
ap.add_argument("-p", "--ports-start", required=True, help="the start of the range of ports to use", type=int)
ap.add_argument("-t", "--host", default="None", help="the host; default is local host; string either internet domain or IPv4", type=str)
args = vars(ap.parse_args())
number_servers = args["number_servers"]
ports_start = args["ports_start"]
host = args["host"]
if host == 'None':
host = socket.gethostname()
dump = 100
example_environment = resume_env(plot=False, step=100, dump=dump)
use_best_model = True
environments = []
for crrt_simu in range(number_servers):
environments.append(RemoteEnvironmentClient(
example_environment, verbose=0, port=ports_start + crrt_simu, host=host,
timing_print=(crrt_simu == 0)
))
if use_best_model:
evaluation_environment = environments.pop()
else:
evaluation_environment = None
import os
import socket
import numpy as np
import csv
from tensorforce.agents import Agent
from tensorforce.execution import ParallelRunner
from simulation_base.env import resume_env, nb_actuations
example_environment = resume_env(plot=False, dump=100, single_run=True)
deterministic = True
network = [dict(type='dense', size=512), dict(type='dense', size=512)]
saver_restore = dict(directory=os.getcwd() + "/saver_data/", load="best-model")
agent = Agent.create(
# Agent + Environment
agent='ppo',
environment=example_environment,
max_episode_timesteps=nb_actuations,
# TODO: nb_actuations could be specified by Environment.max_episode_timesteps() if it makes sense...
# Network
network=network,
# Optimization
batch_size=40,
learning_rate=1e-3,
subsampling_fraction=0.2,
optimization_steps=25,
import os
import socket
import numpy as np
import csv
from tensorforce.agents import Agent
from tensorforce.execution import ParallelRunner
from simulation_base.env import resume_env, nb_actuations, simulation_duration
example_environment = resume_env(plot=False, single_run=True, dump_debug=1)
deterministic = True
network = [dict(type='dense', size=512), dict(type='dense', size=512)]
saver_restore = dict(directory=os.getcwd() + "/saver_data/")
agent = Agent.create(
# Agent + Environment
agent='ppo',
environment=example_environment,
max_episode_timesteps=nb_actuations,
# TODO: nb_actuations could be specified by Environment.max_episode_timesteps() if it makes sense...
# Network
network=network,
# Optimization
batch_size=20,
learning_rate=1e-3,
subsampling_fraction=0.2,
optimization_steps=25,