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

Merge pull request #7866 from initrv/cleanup-tensorboard-callback

Browse Source 
Cleanup tensorboard callback
This commit is contained in:
Robert Caulk 2022-12-13 09:05:46 +01:00 committed by GitHub
commit e6da646e2f
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
6 changed files with 73 additions and 36 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
docs/freqai-reinforcement-learning.md
View File

@ -247,6 +247,32 @@ where `unique-id` is the `identifier` set in the `freqai` configuration file. Th
![tensorboard](assets/tensorboard.jpg)
### Custom logging
FreqAI also provides a built in episodic summary logger called `self.tensorboard_log` for adding custom information to the Tensorboard log. By default, this function is already called once per step inside the environment to record the agent actions. All values accumulated for all steps in a single episode are reported at the conclusion of each episode, followed by a full reset of all metrics to 0 in preparation for the subsequent episode.
`self.tensorboard_log` can also be used anywhere inside the environment, for example, it can be added to the `calculate_reward` function to collect more detailed information about how often various parts of the reward were called:
```py
class MyRLEnv(Base5ActionRLEnv):
"""
User made custom environment. This class inherits from BaseEnvironment and gym.env.
Users can override any functions from those parent classes. Here is an example
of a user customized `calculate_reward()` function.
"""
def calculate_reward(self, action: int) -> float:
if not self._is_valid(action):
self.tensorboard_log("is_valid")
return -2
```
!!! Note
The `self.tensorboard_log()` function is designed for tracking incremented objects only i.e. events, actions inside the training environment. If the event of interest is a float, the float can be passed as the second argument e.g. `self.tensorboard_log("float_metric1", 0.23)` would add 0.23 to `float_metric`. In this case you can also disable incrementing using `inc=False` parameter.
### Choosing a base environment
FreqAI provides two base environments, `Base4ActionEnvironment` and `Base5ActionEnvironment`. As the names imply, the environments are customized for agents that can select from 4 or 5 actions. In the `Base4ActionEnvironment`, the agent can enter long, enter short, hold neutral, or exit position. Meanwhile, in the `Base5ActionEnvironment`, the agent has the same actions as Base4, but instead of a single exit action, it separates exit long and exit short. The main changes stemming from the environment selection include:

2
freqtrade/freqai/RL/Base4ActionRLEnv.py
View File

@ -46,9 +46,9 @@ class Base4ActionRLEnv(BaseEnvironment):
self._done = True
self._update_unrealized_total_profit()
step_reward = self.calculate_reward(action)
self.total_reward += step_reward
self.tensorboard_log(self.actions._member_names_[action])
trade_type = None
if self.is_tradesignal(action):

1
freqtrade/freqai/RL/Base5ActionRLEnv.py
View File

@ -49,6 +49,7 @@ 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])
trade_type = None
if self.is_tradesignal(action):

42
freqtrade/freqai/RL/BaseEnvironment.py
View File

@ -2,7 +2,7 @@ import logging
import random
from abc import abstractmethod
from enum import Enum
from typing import Optional, Type
from typing import Optional, Type, Union
import gym
import numpy as np
@ -78,7 +78,7 @@ class BaseEnvironment(gym.Env):
# set here to default 5Ac, but all children envs can override this
self.actions: Type[Enum] = BaseActions
self.custom_info: dict = {}
self.tensorboard_metrics: dict = {}
self.live: bool = False
if dp:
self.live = dp.runmode in (RunMode.DRY_RUN, RunMode.LIVE)
@ -139,20 +139,38 @@ 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):
"""
Function builds the tensorboard_metrics dictionary
to be parsed by the TensorboardCallback. This
function is designed for tracking incremented objects,
events, actions inside the training environment.
For example, a user can call this to track the
frequency of occurence of an `is_valid` call in
their `calculate_reward()`:
def calculate_reward(self, action: int) -> float:
if not self._is_valid(action):
self.tensorboard_log("is_valid")
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
"""
if not inc or metric not in self.tensorboard_metrics:
self.tensorboard_metrics[metric] = value
else:
self.tensorboard_metrics[metric] += value
def reset_tensorboard_log(self):
self.tensorboard_metrics = {}
def reset(self):
"""
Reset is called at the beginning of every episode
"""
# custom_info is used for episodic reports and tensorboard logging
self.custom_info["Invalid"] = 0
self.custom_info["Hold"] = 0
self.custom_info["Unknown"] = 0
self.custom_info["pnl_factor"] = 0
self.custom_info["duration_factor"] = 0
self.custom_info["reward_exit"] = 0
self.custom_info["reward_hold"] = 0
for action in self.actions:
self.custom_info[f"{action.name}"] = 0
self.reset_tensorboard_log()
self._done = False

29
freqtrade/freqai/RL/TensorboardCallback.py
View File

@ -42,19 +42,18 @@ class TensorboardCallback(BaseCallback):
)
def _on_step(self) -> bool:
custom_info = self.training_env.get_attr("custom_info")[0]
self.logger.record("_state/position", self.locals["infos"][0]["position"])
self.logger.record("_state/trade_duration", self.locals["infos"][0]["trade_duration"])
self.logger.record("_state/current_profit_pct", self.locals["infos"]
[0]["current_profit_pct"])
self.logger.record("_reward/total_profit", self.locals["infos"][0]["total_profit"])
self.logger.record("_reward/total_reward", self.locals["infos"][0]["total_reward"])
self.logger.record_mean("_reward/mean_trade_duration", self.locals["infos"]
[0]["trade_duration"])
self.logger.record("_actions/action", self.locals["infos"][0]["action"])
self.logger.record("_actions/_Invalid", custom_info["Invalid"])
self.logger.record("_actions/_Unknown", custom_info["Unknown"])
self.logger.record("_actions/Hold", custom_info["Hold"])
for action in self.actions:
self.logger.record(f"_actions/{action.name}", custom_info[action.name])
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 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])
return True

9
freqtrade/freqai/prediction_models/ReinforcementLearner.py
View File

@ -100,7 +100,7 @@ class ReinforcementLearner(BaseReinforcementLearningModel):
"""
# first, penalize if the action is not valid
if not self._is_valid(action):
self.custom_info["Invalid"] += 1
self.tensorboard_log("is_valid")
return -2
pnl = self.get_unrealized_profit()
@ -109,15 +109,12 @@ class ReinforcementLearner(BaseReinforcementLearningModel):
# reward agent for entering trades
if (action == Actions.Long_enter.value
and self._position == Positions.Neutral):
self.custom_info[f"{Actions.Long_enter.name}"] += 1
return 25
if (action == Actions.Short_enter.value
and self._position == Positions.Neutral):
self.custom_info[f"{Actions.Short_enter.name}"] += 1
return 25
# discourage agent from not entering trades
if action == Actions.Neutral.value and self._position == Positions.Neutral:
self.custom_info[f"{Actions.Neutral.name}"] += 1
return -1
max_trade_duration = self.rl_config.get('max_trade_duration_candles', 300)
@ -131,22 +128,18 @@ class ReinforcementLearner(BaseReinforcementLearningModel):
# discourage sitting in position
if (self._position in (Positions.Short, Positions.Long) and
action == Actions.Neutral.value):
self.custom_info["Hold"] += 1
return -1 * trade_duration / max_trade_duration
# close long
if action == Actions.Long_exit.value and self._position == Positions.Long:
if pnl > self.profit_aim * self.rr:
factor *= self.rl_config['model_reward_parameters'].get('win_reward_factor', 2)
self.custom_info[f"{Actions.Long_exit.name}"] += 1
return float(pnl * factor)
# close short
if action == Actions.Short_exit.value and self._position == Positions.Short:
if pnl > self.profit_aim * self.rr:
factor *= self.rl_config['model_reward_parameters'].get('win_reward_factor', 2)
self.custom_info[f"{Actions.Short_exit.name}"] += 1
return float(pnl * factor)
self.custom_info["Unknown"] += 1
return 0.