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

improve price df handling to enable backtesting

Browse Source
This commit is contained in:
robcaulk
2022-08-17 12:51:14 +02:00
parent 2080ff86ed
commit b90da46b1b
8 changed files with 77 additions and 59 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -3,9 +3,8 @@ from typing import Any, Dict # , Tuple
import numpy as np
# import numpy.typing as npt
# import pandas as pd
import torch as th
# from pandas import DataFrame
from pandas import DataFrame
from stable_baselines3 import PPO
from stable_baselines3.common.callbacks import EvalCallback
from stable_baselines3.common.monitor import Monitor
@@ -22,7 +21,8 @@ class ReinforcementLearningPPO(BaseReinforcementLearningModel):
User created Reinforcement Learning Model prediction model.
"""
def fit_rl(self, data_dictionary: Dict[str, Any], pair: str, dk: FreqaiDataKitchen):
def fit_rl(self, data_dictionary: Dict[str, Any], pair: str, dk: FreqaiDataKitchen,
prices_train: DataFrame, prices_test: DataFrame):
agent_params = self.freqai_info['model_training_parameters']
reward_params = self.freqai_info['model_reward_parameters']
@@ -31,18 +31,12 @@ class ReinforcementLearningPPO(BaseReinforcementLearningModel):
eval_freq = agent_params.get("eval_cycles", 4) * len(test_df)
total_timesteps = agent_params["train_cycles"] * len(train_df)
# price data for model training and evaluation
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(
len(test_df.index))
# environments
train_env = MyRLEnv(df=train_df, prices=price, window_size=self.CONV_WIDTH,
train_env = MyRLEnv(df=train_df, prices=prices_train, window_size=self.CONV_WIDTH,
reward_kwargs=reward_params)
eval = MyRLEnv(df=test_df, prices=price_test,
eval = MyRLEnv(df=test_df, prices=prices_test,
window_size=self.CONV_WIDTH, reward_kwargs=reward_params)
eval_env = Monitor(eval, ".")
eval_env.reset()
path = dk.data_path
eval_callback = EvalCallback(eval_env, best_model_save_path=f"{path}/",
@@ -63,7 +57,7 @@ class ReinforcementLearningPPO(BaseReinforcementLearningModel):
callback=eval_callback
)
best_model = PPO.load(dk.data_path / "best_model.zip")
best_model = PPO.load(dk.data_path / "best_model")
print('Training finished!')

16
freqtrade/freqai/prediction_models/ReinforcementLearningPPO_multiproc.py
View File

@@ -16,6 +16,7 @@ from freqtrade.freqai.RL.Base3ActionRLEnv import Base3ActionRLEnv, Actions, Posi
from freqtrade.freqai.RL.BaseReinforcementLearningModel import BaseReinforcementLearningModel
from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
import gym
from pandas import DataFrame
logger = logging.getLogger(__name__)
@@ -47,7 +48,8 @@ class ReinforcementLearningPPO_multiproc(BaseReinforcementLearningModel):
User created Reinforcement Learning Model prediction model.
"""
def fit_rl(self, data_dictionary: Dict[str, Any], pair: str, dk: FreqaiDataKitchen):
def fit_rl(self, data_dictionary: Dict[str, Any], pair: str, dk: FreqaiDataKitchen,
prices_train: DataFrame, prices_test: DataFrame):
agent_params = self.freqai_info['model_training_parameters']
reward_params = self.freqai_info['model_reward_parameters']
@@ -57,18 +59,14 @@ class ReinforcementLearningPPO_multiproc(BaseReinforcementLearningModel):
total_timesteps = agent_params["train_cycles"] * len(train_df)
learning_rate = agent_params["learning_rate"]
# price data for model training and evaluation
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(
len(test_df.index))
env_id = "train_env"
th.set_num_threads(dk.thread_count)
num_cpu = int(dk.thread_count / 2)
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, prices_train, reward_params,
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,
eval_env = SubprocVecEnv([make_env(eval_env_id, i, 1, test_df, prices_test, reward_params,
self.CONV_WIDTH, monitor=True) for i in range(num_cpu)])
path = dk.data_path
@@ -92,7 +90,7 @@ class ReinforcementLearningPPO_multiproc(BaseReinforcementLearningModel):
callback=eval_callback
)
best_model = PPO.load(dk.data_path / "best_model.zip")
best_model = PPO.load(dk.data_path / "best_model")
print('Training finished!')
eval_env.close()

15
freqtrade/freqai/prediction_models/ReinforcementLearningTDQN.py
View File

@@ -10,6 +10,7 @@ from freqtrade.freqai.RL.TDQNagent import TDQN
from stable_baselines3 import DQN
from stable_baselines3.common.buffers import ReplayBuffer
import numpy as np
from pandas import DataFrame
from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
@@ -21,7 +22,8 @@ class ReinforcementLearningTDQN(BaseReinforcementLearningModel):
User created Reinforcement Learning Model prediction model.
"""
def fit_rl(self, data_dictionary: Dict[str, Any], pair: str, dk: FreqaiDataKitchen):
def fit_rl(self, data_dictionary: Dict[str, Any], pair: str, dk: FreqaiDataKitchen,
prices_train: DataFrame, prices_test: DataFrame):
agent_params = self.freqai_info['model_training_parameters']
reward_params = self.freqai_info['model_reward_parameters']
@@ -30,15 +32,10 @@ class ReinforcementLearningTDQN(BaseReinforcementLearningModel):
eval_freq = agent_params["eval_cycles"] * len(test_df)
total_timesteps = agent_params["train_cycles"] * len(train_df)
# price data for model training and evaluation
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(
len(test_df.index))
# environments
train_env = MyRLEnv(df=train_df, prices=price, window_size=self.CONV_WIDTH,
train_env = MyRLEnv(df=train_df, prices=prices_train, window_size=self.CONV_WIDTH,
reward_kwargs=reward_params)
eval = MyRLEnv(df=test_df, prices=price_test,
eval = MyRLEnv(df=test_df, prices=prices_test,
window_size=self.CONV_WIDTH, reward_kwargs=reward_params)
eval_env = Monitor(eval, ".")
eval_env.reset()
@@ -66,7 +63,7 @@ class ReinforcementLearningTDQN(BaseReinforcementLearningModel):
callback=eval_callback
)
best_model = DQN.load(dk.data_path / "best_model.zip")
best_model = DQN.load(dk.data_path / "best_model")
print('Training finished!')

14
freqtrade/freqai/prediction_models/ReinforcementLearningTDQN_multiproc.py
View File

@@ -15,7 +15,7 @@ from freqtrade.freqai.RL.BaseReinforcementLearningModel import BaseReinforcement
from freqtrade.freqai.RL.TDQNagent import TDQN
from stable_baselines3.common.buffers import ReplayBuffer
from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
from pandas import DataFrame
logger = logging.getLogger(__name__)
@@ -47,7 +47,8 @@ class ReinforcementLearningTDQN_multiproc(BaseReinforcementLearningModel):
User created Reinforcement Learning Model prediction model.
"""
def fit_rl(self, data_dictionary: Dict[str, Any], pair: str, dk: FreqaiDataKitchen):
def fit_rl(self, data_dictionary: Dict[str, Any], pair: str, dk: FreqaiDataKitchen,
prices_train: DataFrame, prices_test: DataFrame):
agent_params = self.freqai_info['model_training_parameters']
reward_params = self.freqai_info['model_reward_parameters']
@@ -57,18 +58,13 @@ class ReinforcementLearningTDQN_multiproc(BaseReinforcementLearningModel):
total_timesteps = agent_params["train_cycles"] * len(train_df)
learning_rate = agent_params["learning_rate"]
# price data for model training and evaluation
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(
len(test_df.index))
env_id = "train_env"
num_cpu = int(dk.thread_count / 2)
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, prices_train, reward_params,
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,
eval_env = SubprocVecEnv([make_env(eval_env_id, i, 1, test_df, prices_test, reward_params,
self.CONV_WIDTH, monitor=True) for i in range(num_cpu)])
path = dk.data_path