def test_launch_visualizer_canceled(tmpdir):
agent = MockAgent()
gymlike_env = MagicMock()
action_meanings = {0: 'hoge', 1: 'fuga'}
os.makedirs(os.path.join(tmpdir, 'log_space'))
with change_execution_dir(tmpdir):
with patch('chainerrl_visualizer.launcher.input', side_effect='n'), \
patch('chainerrl_visualizer.launcher.inspect_agent') as inspect_agent:
launch_visualizer(agent, gymlike_env, action_meanings)
inspect_agent.assert_not_called()
def test_launch_visualizer(tmpdir, outs):
agent = MockAgent()
agent.model = MagicMock(side_effect=lambda *args: outs)
gymlike_env = MagicMock(spec=gym.Env)
action_meanings = {0: 'hoge', 1: 'fuga'}
# This assertion checks the instances is called correctly.
# In the target luncher function, the instance is called from forked process internally,
# So cannot use `assert_called` and shared values.
# It's not smart, and address with touch file to check called or not.
websrv_called_touch = os.path.join(tmpdir, 'websrv_called.log')
worker_called_touch = os.path.join(tmpdir, 'worker_called.log')
def assert_server_called(*args):
assert len(args) == 12
assert id(args[0]) == id(agent)
assert id(args[1]) == id(gymlike_env)
open(websrv_called_touch, 'w').close()
web_server = MagicMock(side_effect=assert_server_called)
def assert_worker_called(*args):
assert len(args) == 6
assert id(args[0]) == id(agent)
assert id(args[1]) == id(gymlike_env)
open(worker_called_touch, 'w').close()
job_worker = MagicMock(side_effect=assert_worker_called)
webbrowser = MagicMock()
webbrowser.open_new_tab = MagicMock()
with change_execution_dir(tmpdir):
with patch('chainerrl_visualizer.launcher.modify_gym_env_render') as modify_gymenv, \
patch('chainerrl_visualizer.launcher.web_server', web_server), \
patch('chainerrl_visualizer.launcher.job_worker', job_worker), \
patch('chainerrl_visualizer.launcher.webbrowser', webbrowser):
launch_visualizer(agent, gymlike_env, action_meanings)
modify_gymenv.assert_called_once()
assert os.path.exists(websrv_called_touch)
assert os.path.exists(worker_called_touch)
buy_sell_num_flag = [
1.0, 0.0, abs(buy_sell_count)
] if buy_sell_count >= 1 else [0.0, 1.0, abs(buy_sell_count)]
agent.stop_episode_and_train(X_train[-1] + buy_sell_num_flag, reward,
True) #エピソード（価格データの最初から最後までを辿ること）を一旦停止
#強化学習の初期では少ない手数料で、学習後半には手数料を少しずつ増やしていく
print("========Training END============")
print("Passは" + str(pass_count) + "回")
print("終わった後のbuy_sell_count:" + str(buy_sell_count) +
("回買いの取引が多い" if buy_sell_count > 0 else "回売りの取引が多い"))
print("Initial MONEY" + str(first_total_money))
print("FINAL MONEY:" + str(total_money))
# Save an agent to the 'agent' directory
agent.save('chainerRLAgentFinal-Dense')
'''
ACTION_MEANINGS = {
0: 'a',
1: 'b',
2: 'c'
}
launch_visualizer(
agent, # required
env, # required
ACTION_MEANINGS, # required
port=5002, # optional (default: 5002)
log_dir='log_space', # optional (default: 'log_space')
raw_image_input=False, # optional (default: False)
contains_rnn=False, # optional (default: False)
)
target_update_interval=10**4,
update_interval=4,
batch_accumulator="mean",
phi=phi,
)
agent.load("parameters")
ACTION_MEANINGS = {
0: "NOOP",
1: "FIRE",
2: "UP",
3: "RIGHT",
4: "LEFT",
5: "DOWN",
6: "UPRIGHT",
7: "UPLEFT",
8: "DOWNRIGHT",
9: "DOWNLEFT",
10: "UPFIRE",
11: "RIGHTFIRE",
12: "LEFTFIRE",
13: "DOWNFIRE",
14: "UPRIGHTFIRE",
15: "UPLEFTFIRE",
16: "DOWNRIGHTFIRE",
17: "DOWNLEFTFIRE",
}
launch_visualizer(agent, env, ACTION_MEANINGS, raw_image_input=True)
action_space = env.action_space
model = A3CFFGaussian(obs_space.low.size,
action_space,
bound_mean=False,
normalize_obs=False)
opt = chainer.optimizers.Adam(alpha=3e-4, eps=1e-5)
opt.setup(model)
agent = PPO(model,
opt,
gpu=-1,
phi=phi,
update_interval=2048,
minibatch_size=64,
epochs=10,
clip_eps_vf=None,
entropy_coef=0.0,
standardize_advantages=False)
agent.load("parameters")
ACTION_MEANINGS = {
0: 'Hip1(Torque/Velocity)',
1: 'Knee1(Torque/Velocity)',
2: 'Hip2(Torque/Velocity)',
3: 'Knee2(Torque/Velocity)',
}
launch_visualizer(agent, env, ACTION_MEANINGS)