import os
import sys
current_path = os.getcwd()
sys.path.append(current_path)
from hypertune import start_commander, start_workers
from coordinator import Coordinator
from config import mode, number_workers, tuned_config, fix_random_seed
import numpy as np
if fix_random_seed:
np.random.seed(123)
# %matplotlib inline
if mode == 'parallel':
start_commander()
workers = start_workers(number_workers)
else:
model = Coordinator(tuned_config, '2900')
#################### to do ####################
# model = Coordinator(tuned_config, '-5.28')
# model.restore_price_predictor('-5.28-80000-')
##############################################
model.train('single', True)
model.back_test('test', 2500, True)
ob = env.reset()
for i in range(5):
print(coo.action_values(ob))
ob, a, r, ob_ = env.step(np.ones(5))
coo.train(env,
total_training_step=total_training_step,
replay_period=replay_period,
tensorboard=True)
env_test = PortfolioEnv(df_train,
steps=2500,
trading_cost=0.0,
window_length=window_length,
scale=False,
random_reset=False)
ob = env_test.reset()
for i in range(10):
print(coo.action_values(ob))
print(np.argmax(coo.action_values(ob)))
ob, a, r, ob_ = env.step(np.ones(5))
# coo.restore('')
# l = coo.network_state()
# print(l['training_network/output/weights:0'])
coo.back_test(env_test, render_mode='usual')
# coo.open_tb(port='8009')
