Merge remote-tracking branch 'origin/develop' into feat/add-pytorch-model-support
This commit is contained in:
@@ -47,7 +47,7 @@ class Base3ActionRLEnv(BaseEnvironment):
|
||||
self._update_unrealized_total_profit()
|
||||
step_reward = self.calculate_reward(action)
|
||||
self.total_reward += step_reward
|
||||
self.tensorboard_log(self.actions._member_names_[action])
|
||||
self.tensorboard_log(self.actions._member_names_[action], category="actions")
|
||||
|
||||
trade_type = None
|
||||
if self.is_tradesignal(action):
|
||||
@@ -66,7 +66,7 @@ class Base3ActionRLEnv(BaseEnvironment):
|
||||
elif action == Actions.Sell.value and not self.can_short:
|
||||
self._update_total_profit()
|
||||
self._position = Positions.Neutral
|
||||
trade_type = "neutral"
|
||||
trade_type = "exit"
|
||||
self._last_trade_tick = None
|
||||
else:
|
||||
print("case not defined")
|
||||
@@ -74,7 +74,7 @@ class Base3ActionRLEnv(BaseEnvironment):
|
||||
if trade_type is not None:
|
||||
self.trade_history.append(
|
||||
{'price': self.current_price(), 'index': self._current_tick,
|
||||
'type': trade_type})
|
||||
'type': trade_type, 'profit': self.get_unrealized_profit()})
|
||||
|
||||
if (self._total_profit < self.max_drawdown or
|
||||
self._total_unrealized_profit < self.max_drawdown):
|
||||
|
@@ -48,20 +48,10 @@ class Base4ActionRLEnv(BaseEnvironment):
|
||||
self._update_unrealized_total_profit()
|
||||
step_reward = self.calculate_reward(action)
|
||||
self.total_reward += step_reward
|
||||
self.tensorboard_log(self.actions._member_names_[action])
|
||||
self.tensorboard_log(self.actions._member_names_[action], category="actions")
|
||||
|
||||
trade_type = None
|
||||
if self.is_tradesignal(action):
|
||||
"""
|
||||
Action: Neutral, position: Long -> Close Long
|
||||
Action: Neutral, position: Short -> Close Short
|
||||
|
||||
Action: Long, position: Neutral -> Open Long
|
||||
Action: Long, position: Short -> Close Short and Open Long
|
||||
|
||||
Action: Short, position: Neutral -> Open Short
|
||||
Action: Short, position: Long -> Close Long and Open Short
|
||||
"""
|
||||
|
||||
if action == Actions.Neutral.value:
|
||||
self._position = Positions.Neutral
|
||||
@@ -69,16 +59,16 @@ class Base4ActionRLEnv(BaseEnvironment):
|
||||
self._last_trade_tick = None
|
||||
elif action == Actions.Long_enter.value:
|
||||
self._position = Positions.Long
|
||||
trade_type = "long"
|
||||
trade_type = "enter_long"
|
||||
self._last_trade_tick = self._current_tick
|
||||
elif action == Actions.Short_enter.value:
|
||||
self._position = Positions.Short
|
||||
trade_type = "short"
|
||||
trade_type = "enter_short"
|
||||
self._last_trade_tick = self._current_tick
|
||||
elif action == Actions.Exit.value:
|
||||
self._update_total_profit()
|
||||
self._position = Positions.Neutral
|
||||
trade_type = "neutral"
|
||||
trade_type = "exit"
|
||||
self._last_trade_tick = None
|
||||
else:
|
||||
print("case not defined")
|
||||
@@ -86,7 +76,7 @@ class Base4ActionRLEnv(BaseEnvironment):
|
||||
if trade_type is not None:
|
||||
self.trade_history.append(
|
||||
{'price': self.current_price(), 'index': self._current_tick,
|
||||
'type': trade_type})
|
||||
'type': trade_type, 'profit': self.get_unrealized_profit()})
|
||||
|
||||
if (self._total_profit < self.max_drawdown or
|
||||
self._total_unrealized_profit < self.max_drawdown):
|
||||
|
@@ -49,20 +49,10 @@ class Base5ActionRLEnv(BaseEnvironment):
|
||||
self._update_unrealized_total_profit()
|
||||
step_reward = self.calculate_reward(action)
|
||||
self.total_reward += step_reward
|
||||
self.tensorboard_log(self.actions._member_names_[action])
|
||||
self.tensorboard_log(self.actions._member_names_[action], category="actions")
|
||||
|
||||
trade_type = None
|
||||
if self.is_tradesignal(action):
|
||||
"""
|
||||
Action: Neutral, position: Long -> Close Long
|
||||
Action: Neutral, position: Short -> Close Short
|
||||
|
||||
Action: Long, position: Neutral -> Open Long
|
||||
Action: Long, position: Short -> Close Short and Open Long
|
||||
|
||||
Action: Short, position: Neutral -> Open Short
|
||||
Action: Short, position: Long -> Close Long and Open Short
|
||||
"""
|
||||
|
||||
if action == Actions.Neutral.value:
|
||||
self._position = Positions.Neutral
|
||||
@@ -70,21 +60,21 @@ class Base5ActionRLEnv(BaseEnvironment):
|
||||
self._last_trade_tick = None
|
||||
elif action == Actions.Long_enter.value:
|
||||
self._position = Positions.Long
|
||||
trade_type = "long"
|
||||
trade_type = "enter_long"
|
||||
self._last_trade_tick = self._current_tick
|
||||
elif action == Actions.Short_enter.value:
|
||||
self._position = Positions.Short
|
||||
trade_type = "short"
|
||||
trade_type = "enter_short"
|
||||
self._last_trade_tick = self._current_tick
|
||||
elif action == Actions.Long_exit.value:
|
||||
self._update_total_profit()
|
||||
self._position = Positions.Neutral
|
||||
trade_type = "neutral"
|
||||
trade_type = "exit_long"
|
||||
self._last_trade_tick = None
|
||||
elif action == Actions.Short_exit.value:
|
||||
self._update_total_profit()
|
||||
self._position = Positions.Neutral
|
||||
trade_type = "neutral"
|
||||
trade_type = "exit_short"
|
||||
self._last_trade_tick = None
|
||||
else:
|
||||
print("case not defined")
|
||||
@@ -92,7 +82,7 @@ class Base5ActionRLEnv(BaseEnvironment):
|
||||
if trade_type is not None:
|
||||
self.trade_history.append(
|
||||
{'price': self.current_price(), 'index': self._current_tick,
|
||||
'type': trade_type})
|
||||
'type': trade_type, 'profit': self.get_unrealized_profit()})
|
||||
|
||||
if (self._total_profit < self.max_drawdown or
|
||||
self._total_unrealized_profit < self.max_drawdown):
|
||||
|
@@ -137,7 +137,8 @@ class BaseEnvironment(gym.Env):
|
||||
self.np_random, seed = seeding.np_random(seed)
|
||||
return [seed]
|
||||
|
||||
def tensorboard_log(self, metric: str, value: Union[int, float] = 1, inc: bool = True):
|
||||
def tensorboard_log(self, metric: str, value: Optional[Union[int, float]] = None,
|
||||
inc: Optional[bool] = None, category: str = "custom"):
|
||||
"""
|
||||
Function builds the tensorboard_metrics dictionary
|
||||
to be parsed by the TensorboardCallback. This
|
||||
@@ -149,17 +150,24 @@ class BaseEnvironment(gym.Env):
|
||||
|
||||
def calculate_reward(self, action: int) -> float:
|
||||
if not self._is_valid(action):
|
||||
self.tensorboard_log("is_valid")
|
||||
self.tensorboard_log("invalid")
|
||||
return -2
|
||||
|
||||
:param metric: metric to be tracked and incremented
|
||||
:param value: value to increment `metric` by
|
||||
:param inc: sets whether the `value` is incremented or not
|
||||
:param value: `metric` value
|
||||
:param inc: (deprecated) sets whether the `value` is incremented or not
|
||||
:param category: `metric` category
|
||||
"""
|
||||
if not inc or metric not in self.tensorboard_metrics:
|
||||
self.tensorboard_metrics[metric] = value
|
||||
increment = True if value is None else False
|
||||
value = 1 if increment else value
|
||||
|
||||
if category not in self.tensorboard_metrics:
|
||||
self.tensorboard_metrics[category] = {}
|
||||
|
||||
if not increment or metric not in self.tensorboard_metrics[category]:
|
||||
self.tensorboard_metrics[category][metric] = value
|
||||
else:
|
||||
self.tensorboard_metrics[metric] += value
|
||||
self.tensorboard_metrics[category][metric] += value
|
||||
|
||||
def reset_tensorboard_log(self):
|
||||
self.tensorboard_metrics = {}
|
||||
|
@@ -114,6 +114,7 @@ class BaseReinforcementLearningModel(IFreqaiModel):
|
||||
|
||||
# normalize all data based on train_dataset only
|
||||
prices_train, prices_test = self.build_ohlc_price_dataframes(dk.data_dictionary, pair, dk)
|
||||
|
||||
data_dictionary = dk.normalize_data(data_dictionary)
|
||||
|
||||
# data cleaning/analysis
|
||||
@@ -148,12 +149,8 @@ class BaseReinforcementLearningModel(IFreqaiModel):
|
||||
|
||||
env_info = self.pack_env_dict(dk.pair)
|
||||
|
||||
self.train_env = self.MyRLEnv(df=train_df,
|
||||
prices=prices_train,
|
||||
**env_info)
|
||||
self.eval_env = Monitor(self.MyRLEnv(df=test_df,
|
||||
prices=prices_test,
|
||||
**env_info))
|
||||
self.train_env = self.MyRLEnv(df=train_df, prices=prices_train, **env_info)
|
||||
self.eval_env = Monitor(self.MyRLEnv(df=test_df, prices=prices_test, **env_info))
|
||||
self.eval_callback = EvalCallback(self.eval_env, deterministic=True,
|
||||
render=False, eval_freq=len(train_df),
|
||||
best_model_save_path=str(dk.data_path))
|
||||
@@ -238,6 +235,9 @@ class BaseReinforcementLearningModel(IFreqaiModel):
|
||||
filtered_dataframe, _ = dk.filter_features(
|
||||
unfiltered_df, dk.training_features_list, training_filter=False
|
||||
)
|
||||
|
||||
filtered_dataframe = self.drop_ohlc_from_df(filtered_dataframe, dk)
|
||||
|
||||
filtered_dataframe = dk.normalize_data_from_metadata(filtered_dataframe)
|
||||
dk.data_dictionary["prediction_features"] = filtered_dataframe
|
||||
|
||||
@@ -285,7 +285,6 @@ class BaseReinforcementLearningModel(IFreqaiModel):
|
||||
train_df = data_dictionary["train_features"]
|
||||
test_df = data_dictionary["test_features"]
|
||||
|
||||
# %-raw_volume_gen_shift-2_ETH/USDT_1h
|
||||
# price data for model training and evaluation
|
||||
tf = self.config['timeframe']
|
||||
rename_dict = {'%-raw_open': 'open', '%-raw_low': 'low',
|
||||
@@ -318,8 +317,24 @@ class BaseReinforcementLearningModel(IFreqaiModel):
|
||||
prices_test.rename(columns=rename_dict, inplace=True)
|
||||
prices_test.reset_index(drop=True)
|
||||
|
||||
train_df = self.drop_ohlc_from_df(train_df, dk)
|
||||
test_df = self.drop_ohlc_from_df(test_df, dk)
|
||||
|
||||
return prices_train, prices_test
|
||||
|
||||
def drop_ohlc_from_df(self, df: DataFrame, dk: FreqaiDataKitchen):
|
||||
"""
|
||||
Given a dataframe, drop the ohlc data
|
||||
"""
|
||||
drop_list = ['%-raw_open', '%-raw_low', '%-raw_high', '%-raw_close']
|
||||
|
||||
if self.rl_config["drop_ohlc_from_features"]:
|
||||
df.drop(drop_list, axis=1, inplace=True)
|
||||
feature_list = dk.training_features_list
|
||||
dk.training_features_list = [e for e in feature_list if e not in drop_list]
|
||||
|
||||
return df
|
||||
|
||||
def load_model_from_disk(self, dk: FreqaiDataKitchen) -> Any:
|
||||
"""
|
||||
Can be used by user if they are trying to limit_ram_usage *and*
|
||||
|
@@ -13,7 +13,7 @@ class TensorboardCallback(BaseCallback):
|
||||
episodic summary reports.
|
||||
"""
|
||||
def __init__(self, verbose=1, actions: Type[Enum] = BaseActions):
|
||||
super(TensorboardCallback, self).__init__(verbose)
|
||||
super().__init__(verbose)
|
||||
self.model: Any = None
|
||||
self.logger = None # type: Any
|
||||
self.training_env: BaseEnvironment = None # type: ignore
|
||||
@@ -46,14 +46,12 @@ class TensorboardCallback(BaseCallback):
|
||||
local_info = self.locals["infos"][0]
|
||||
tensorboard_metrics = self.training_env.get_attr("tensorboard_metrics")[0]
|
||||
|
||||
for info in local_info:
|
||||
if info not in ["episode", "terminal_observation"]:
|
||||
self.logger.record(f"_info/{info}", local_info[info])
|
||||
for metric in local_info:
|
||||
if metric not in ["episode", "terminal_observation"]:
|
||||
self.logger.record(f"info/{metric}", local_info[metric])
|
||||
|
||||
for info in tensorboard_metrics:
|
||||
if info in [action.name for action in self.actions]:
|
||||
self.logger.record(f"_actions/{info}", tensorboard_metrics[info])
|
||||
else:
|
||||
self.logger.record(f"_custom/{info}", tensorboard_metrics[info])
|
||||
for category in tensorboard_metrics:
|
||||
for metric in tensorboard_metrics[category]:
|
||||
self.logger.record(f"{category}/{metric}", tensorboard_metrics[category][metric])
|
||||
|
||||
return True
|
||||
|
@@ -251,7 +251,7 @@ class FreqaiDataKitchen:
|
||||
(drop_index == 0) & (drop_index_labels == 0)
|
||||
]
|
||||
logger.info(
|
||||
f"dropped {len(unfiltered_df) - len(filtered_df)} training points"
|
||||
f"{self.pair}: dropped {len(unfiltered_df) - len(filtered_df)} training points"
|
||||
f" due to NaNs in populated dataset {len(unfiltered_df)}."
|
||||
)
|
||||
if (1 - len(filtered_df) / len(unfiltered_df)) > 0.1 and self.live:
|
||||
@@ -675,7 +675,7 @@ class FreqaiDataKitchen:
|
||||
]
|
||||
|
||||
logger.info(
|
||||
f"SVM tossed {len(y_pred) - kept_points.sum()}"
|
||||
f"{self.pair}: SVM tossed {len(y_pred) - kept_points.sum()}"
|
||||
f" test points from {len(y_pred)} total points."
|
||||
)
|
||||
|
||||
@@ -949,7 +949,7 @@ class FreqaiDataKitchen:
|
||||
|
||||
if (len(do_predict) - do_predict.sum()) > 0:
|
||||
logger.info(
|
||||
f"DI tossed {len(do_predict) - do_predict.sum()} predictions for "
|
||||
f"{self.pair}: DI tossed {len(do_predict) - do_predict.sum()} predictions for "
|
||||
"being too far from training data."
|
||||
)
|
||||
|
||||
|
@@ -105,6 +105,10 @@ class IFreqaiModel(ABC):
|
||||
self.data_provider: Optional[DataProvider] = None
|
||||
self.max_system_threads = max(int(psutil.cpu_count() * 2 - 2), 1)
|
||||
self.can_short = True # overridden in start() with strategy.can_short
|
||||
self.model: Any = None
|
||||
if self.ft_params.get('principal_component_analysis', False) and self.continual_learning:
|
||||
self.ft_params.update({'principal_component_analysis': False})
|
||||
logger.warning('User tried to use PCA with continual learning. Deactivating PCA.')
|
||||
|
||||
record_params(config, self.full_path)
|
||||
|
||||
@@ -154,8 +158,7 @@ class IFreqaiModel(ABC):
|
||||
dk = self.start_backtesting(dataframe, metadata, self.dk, strategy)
|
||||
dataframe = dk.remove_features_from_df(dk.return_dataframe)
|
||||
else:
|
||||
logger.info(
|
||||
"Backtesting using historic predictions (live models)")
|
||||
logger.info("Backtesting using historic predictions (live models)")
|
||||
dk = self.start_backtesting_from_historic_predictions(
|
||||
dataframe, metadata, self.dk)
|
||||
dataframe = dk.return_dataframe
|
||||
@@ -339,13 +342,14 @@ class IFreqaiModel(ABC):
|
||||
except Exception as msg:
|
||||
logger.warning(
|
||||
f"Training {pair} raised exception {msg.__class__.__name__}. "
|
||||
f"Message: {msg}, skipping.")
|
||||
f"Message: {msg}, skipping.", exc_info=True)
|
||||
self.model = None
|
||||
|
||||
self.dd.pair_dict[pair]["trained_timestamp"] = int(
|
||||
tr_train.stopts)
|
||||
if self.plot_features:
|
||||
if self.plot_features and self.model is not None:
|
||||
plot_feature_importance(self.model, pair, dk, self.plot_features)
|
||||
if self.save_backtest_models:
|
||||
if self.save_backtest_models and self.model is not None:
|
||||
logger.info('Saving backtest model to disk.')
|
||||
self.dd.save_data(self.model, pair, dk)
|
||||
else:
|
||||
|
@@ -100,7 +100,7 @@ class ReinforcementLearner(BaseReinforcementLearningModel):
|
||||
"""
|
||||
# first, penalize if the action is not valid
|
||||
if not self._is_valid(action):
|
||||
self.tensorboard_log("is_valid")
|
||||
self.tensorboard_log("invalid", category="actions")
|
||||
return -2
|
||||
|
||||
pnl = self.get_unrealized_profit()
|
||||
|
Reference in New Issue
Block a user