def main():
#env_id = "CartPole-v1"
vix_env = trading_vix_env.trading_vix_env()
num_cpu = 20 # Number of processes to use
# Create the vectorized environment
env = SubprocVecEnv([make_env(vix_env, i) for i in range(num_cpu)])
model = A2C('MlpPolicy', env, verbose=1, n_steps=5)
model.learn(total_timesteps=2500000000)
def main():
#env_id = "CartPole-v1"
vix_env = trading_vix_env.trading_vix_env()
num_cpu = 20 # Number of processes to use
# Create the vectorized environment
env = SubprocVecEnv([make_env(vix_env, i) for i in range(num_cpu)])
# Create log dir
log_dir = './ppo_data'
os.makedirs(log_dir, exist_ok=True)
env = VecMonitor(env, log_dir)
callback = custom_call_back.CustomCallback(check_freq = 1000,log_dir = log_dir)
model = PPO('MlpPolicy', env, verbose=1,n_steps=500,batch_size = 10000)
model.learn(total_timesteps=2500000000,callback = callback)
def main():
# Create log dir
log_dir = './ddpg_data'
os.makedirs(log_dir, exist_ok=True)
vix_env = trading_vix_env.trading_vix_env()
env = Monitor(vix_env, log_dir)
# Create action noise because TD3 and DDPG use a deterministic policy
n_actions = env.action_space.shape[-1]
action_noise = NormalActionNoise(mean=np.zeros(n_actions), sigma=0.1 * np.ones(n_actions))
# Create the callback: check every 20000 steps
callback = custom_call_back.CustomCallback(check_freq = 20000,log_dir = log_dir)
# Create RL model
model = DDPG('MlpPolicy',env,action_noise = action_noise, verbose=2,batch_size = 10000)
# Train the agent
model.learn(total_timesteps=int(5e9), callback=callback)
import config as C
import numpy as np
import trading_vix
import trading_vix_env
import utils
import matplotlib.pyplot as plt
import jsonpickle
import train
import os
if __name__ == '__main__':
np.random.seed(1234)
#env = trading_vix.trading_vix()
env = trading_vix_env.trading_vix_env()
theta, episode_rewards = train.train(N=C.batch_size,
T=C.training_epoch,
delta=1e-2,
env=env)
'''
param N: number of trajectories to sample in each time step
param T: number of iterations to train the model
param delta: trust region size, because we are using trpo
param env: the environment for the policy to learn
'''
# #test the training result
# observation = env.reset()
# current_feature = utils.extract_features(observation,C.output_dim)
# for t in range(200):