adelorenzo/stable
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

learning_rate + multicpu changes

Browse Source
This commit is contained in:
MukavaValkku 2022-08-15 18:35:41 +03:00 committed by robcaulk
parent 48bb51b458
commit 57c488a6f1
1 changed files with 11 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
freqtrade/freqai/prediction_models/ReinforcementLearningPPO_multiproc.py
View File

@ -50,19 +50,22 @@ class ReinforcementLearningPPO_multiproc(BaseReinforcementLearningModel):
test_df = data_dictionary["test_features"] test_df = data_dictionary["test_features"]
eval_freq = agent_params.get("eval_cycles", 4) * len(test_df) eval_freq = agent_params.get("eval_cycles", 4) * len(test_df)
total_timesteps = agent_params["train_cycles"] * len(train_df) total_timesteps = agent_params["train_cycles"] * len(train_df)
learning_rate = agent_params["learning_rate"]
# price data for model training and evaluation # price data for model training and evaluation
price = self.dd.historic_data[pair][f"{self.config['timeframe']}"].tail(len(train_df.index)) price = self.dd.historic_data[pair][f"{self.config['timeframe']}"].tail(len(train_df.index))
price_test = self.dd.historic_data[pair][f"{self.config['timeframe']}"].tail( price_test = self.dd.historic_data[pair][f"{self.config['timeframe']}"].tail(
len(test_df.index)) len(test_df.index))
env_id = "CartPole-v1" env_id = "train_env"
num_cpu = 4 train_num_cpu = 6
train_env = SubprocVecEnv([make_env(env_id, i, 1, train_df, price, reward_params, train_env = SubprocVecEnv([make_env(env_id, i, 1, train_df, price, reward_params,
self.CONV_WIDTH) for i in range(num_cpu)]) self.CONV_WIDTH) for i in range(train_num_cpu)])
eval_env = SubprocVecEnv([make_env(env_id, i, 1, test_df, price_test, reward_params, eval_num_cpu = 6
self.CONV_WIDTH) for i in range(num_cpu)]) eval_env_id = 'eval_env'
eval_env = SubprocVecEnv([make_env(eval_env_id, i, 1, test_df, price_test, reward_params,
self.CONV_WIDTH) for i in range(eval_num_cpu)])
path = self.dk.data_path path = self.dk.data_path
eval_callback = EvalCallback(eval_env, best_model_save_path=f"{path}/", eval_callback = EvalCallback(eval_env, best_model_save_path=f"{path}/",
@ -71,10 +74,10 @@ class ReinforcementLearningPPO_multiproc(BaseReinforcementLearningModel):
# model arch # model arch
policy_kwargs = dict(activation_fn=th.nn.ReLU, policy_kwargs = dict(activation_fn=th.nn.ReLU,
net_arch=[256, 256, 128]) net_arch=[512, 512, 512])
model = PPO('MlpPolicy', train_env, policy_kwargs=policy_kwargs, model = PPO('MlpPolicy', train_env, policy_kwargs=policy_kwargs,
tensorboard_log=f"{path}/ppo/tensorboard/", learning_rate=0.00025, gamma=0.9, verbose=1 tensorboard_log=f"{path}/ppo/tensorboard/", learning_rate=learning_rate, gamma=0.9, verbose=1
) )
model.learn( model.learn(
@ -83,6 +86,7 @@ class ReinforcementLearningPPO_multiproc(BaseReinforcementLearningModel):
) )
print('Training finished!') print('Training finished!')
eval_env.close()
return model return model