oxbrbtx/PrimAITE
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'origin/dev' into feature/1572-fix-docs-formatting

Browse Source
This commit is contained in:
Marek Wolan
2023-07-07 10:30:11 +01:00
parent f4b98542b6 036e0fe342
commit 7e0eee5d73
17 changed files with 590 additions and 200 deletions
Download Patch File Download Diff File Expand all files Collapse all files

11
src/primaite/agents/agent.py
View File

@@ -266,10 +266,19 @@ class AgentSessionABC(ABC):
raise FileNotFoundError(msg)
pass
@property
def _saved_agent_path(self) -> Path:
file_name = (
f"{self._training_config.agent_framework}_"
f"{self._training_config.agent_identifier}_"
f"{self.timestamp_str}.zip"
)
return self.learning_path / file_name
@abstractmethod
def save(self):
"""Save the agent."""
self._agent.save(self.session_path)
pass
@abstractmethod
def export(self):

36
src/primaite/agents/rllib.py
View File

@@ -1,9 +1,11 @@
from __future__ import annotations
import json
import shutil
from datetime import datetime
from pathlib import Path
from typing import Union
from uuid import uuid4
from ray.rllib.algorithms import Algorithm
from ray.rllib.algorithms.a2c import A2CConfig
@@ -118,6 +120,7 @@ class RLlibAgent(AgentSessionABC):
timestamp_str=self.timestamp_str,
),
)
self._agent_config.seed = self._training_config.seed
self._agent_config.training(train_batch_size=self._training_config.num_steps)
self._agent_config.framework(framework="tf")
@@ -132,9 +135,11 @@ class RLlibAgent(AgentSessionABC):
def _save_checkpoint(self):
checkpoint_n = self._training_config.checkpoint_every_n_episodes
episode_count = self._current_result["episodes_total"]
if checkpoint_n > 0 and episode_count > 0:
if (episode_count % checkpoint_n == 0) or (episode_count == self._training_config.num_episodes):
self._agent.save(str(self.checkpoints_path))
save_checkpoint = False
if checkpoint_n:
save_checkpoint = episode_count % checkpoint_n == 0
if episode_count and save_checkpoint:
self._agent.save(str(self.checkpoints_path))
def learn(
self,
@@ -152,9 +157,14 @@ class RLlibAgent(AgentSessionABC):
for i in range(episodes):
self._current_result = self._agent.train()
self._save_checkpoint()
self.save()
self._agent.stop()
super().learn()
# save agent
self.save()
def evaluate(
self,
**kwargs,
@@ -174,9 +184,25 @@ class RLlibAgent(AgentSessionABC):
"""Load an agent from file."""
raise NotImplementedError
def save(self):
def save(self, overwrite_existing: bool = True):
"""Save the agent."""
raise NotImplementedError
# Make temp dir to save in isolation
temp_dir = self.learning_path / str(uuid4())
temp_dir.mkdir()
# Save the agent to the temp dir
self._agent.save(str(temp_dir))
# Capture the saved Rllib checkpoint inside the temp directory
for file in temp_dir.iterdir():
checkpoint_dir = file
break
# Zip the folder
shutil.make_archive(str(self._saved_agent_path).replace(".zip", ""), "zip", checkpoint_dir) # noqa
# Drop the temp directory
shutil.rmtree(temp_dir)
def export(self):
"""Export the agent to transportable file format."""

25
src/primaite/agents/sb3.py
View File

@@ -71,16 +71,19 @@ class SB3Agent(AgentSessionABC):
verbose=self.sb3_output_verbose_level,
n_steps=self._training_config.num_steps,
tensorboard_log=str(self._tensorboard_log_path),
seed=self._training_config.seed,
)
def _save_checkpoint(self):
checkpoint_n = self._training_config.checkpoint_every_n_episodes
episode_count = self._env.episode_count
if checkpoint_n > 0 and episode_count > 0:
if (episode_count % checkpoint_n == 0) or (episode_count == self._training_config.num_episodes):
checkpoint_path = self.checkpoints_path / f"sb3ppo_{episode_count}.zip"
self._agent.save(checkpoint_path)
_LOGGER.debug(f"Saved agent checkpoint: {checkpoint_path}")
save_checkpoint = False
if checkpoint_n:
save_checkpoint = episode_count % checkpoint_n == 0
if episode_count and save_checkpoint:
checkpoint_path = self.checkpoints_path / f"sb3ppo_{episode_count}.zip"
self._agent.save(checkpoint_path)
_LOGGER.debug(f"Saved agent checkpoint: {checkpoint_path}")
def _get_latest_checkpoint(self):
pass
@@ -102,25 +105,27 @@ class SB3Agent(AgentSessionABC):
self._agent.learn(total_timesteps=time_steps)
self._save_checkpoint()
self._env.reset()
self.save()
self._env.close()
super().learn()
# save agent
self.save()
def evaluate(
self,
deterministic: bool = True,
**kwargs,
):
"""
Evaluate the agent.
:param deterministic: Whether the evaluation is deterministic.
:param kwargs: Any agent-specific key-word args to be passed.
"""
time_steps = self._training_config.num_steps
episodes = self._training_config.num_episodes
self._env.set_as_eval()
self.is_eval = True
if deterministic:
if self._training_config.deterministic:
deterministic_str = "deterministic"
else:
deterministic_str = "non-deterministic"
@@ -131,7 +136,7 @@ class SB3Agent(AgentSessionABC):
obs = self._env.reset()
for step in range(time_steps):
action, _states = self._agent.predict(obs, deterministic=deterministic)
action, _states = self._agent.predict(obs, deterministic=self._training_config.deterministic)
if isinstance(action, np.ndarray):
action = np.int64(action)
obs, rewards, done, info = self._env.step(action)
@@ -146,7 +151,7 @@ class SB3Agent(AgentSessionABC):
def save(self):
"""Save the agent."""
raise NotImplementedError
self._agent.save(self._saved_agent_path)
def export(self):
"""Export the agent to transportable file format."""

108
src/primaite/config/_package_data/training/training_config_main.yaml
View File

@@ -31,6 +31,16 @@ agent_identifier: PPO
# False
random_red_agent: False
# The (integer) seed to be used in random number generation
# Default is None (null)
seed: null
# Set whether the agent will be deterministic instead of stochastic
# Options are:
# True
# False
deterministic: False
# Sets what view of the environment the deterministic hardcoded agent has. The default is BASIC.
# Options are:
# "BASIC" (The current observation space only)
@@ -83,58 +93,58 @@ sb3_output_verbose_level: NONE
# Generic
all_ok: 0
# Node Hardware State
off_should_be_on: -10
off_should_be_resetting: -5
on_should_be_off: -2
on_should_be_resetting: -5
resetting_should_be_on: -5
resetting_should_be_off: -2
resetting: -3
off_should_be_on: -0.001
off_should_be_resetting: -0.0005
on_should_be_off: -0.0002
on_should_be_resetting: -0.0005
resetting_should_be_on: -0.0005
resetting_should_be_off: -0.0002
resetting: -0.0003
# Node Software or Service State
good_should_be_patching: 2
good_should_be_compromised: 5
good_should_be_overwhelmed: 5
patching_should_be_good: -5
patching_should_be_compromised: 2
patching_should_be_overwhelmed: 2
patching: -3
compromised_should_be_good: -20
compromised_should_be_patching: -20
compromised_should_be_overwhelmed: -20
compromised: -20
overwhelmed_should_be_good: -20
overwhelmed_should_be_patching: -20
overwhelmed_should_be_compromised: -20
overwhelmed: -20
good_should_be_patching: 0.0002
good_should_be_compromised: 0.0005
good_should_be_overwhelmed: 0.0005
patching_should_be_good: -0.0005
patching_should_be_compromised: 0.0002
patching_should_be_overwhelmed: 0.0002
patching: -0.0003
compromised_should_be_good: -0.002
compromised_should_be_patching: -0.002
compromised_should_be_overwhelmed: -0.002
compromised: -0.002
overwhelmed_should_be_good: -0.002
overwhelmed_should_be_patching: -0.002
overwhelmed_should_be_compromised: -0.002
overwhelmed: -0.002
# Node File System State
good_should_be_repairing: 2
good_should_be_restoring: 2
good_should_be_corrupt: 5
good_should_be_destroyed: 10
repairing_should_be_good: -5
repairing_should_be_restoring: 2
repairing_should_be_corrupt: 2
repairing_should_be_destroyed: 0
repairing: -3
restoring_should_be_good: -10
restoring_should_be_repairing: -2
restoring_should_be_corrupt: 1
restoring_should_be_destroyed: 2
restoring: -6
corrupt_should_be_good: -10
corrupt_should_be_repairing: -10
corrupt_should_be_restoring: -10
corrupt_should_be_destroyed: 2
corrupt: -10
destroyed_should_be_good: -20
destroyed_should_be_repairing: -20
destroyed_should_be_restoring: -20
destroyed_should_be_corrupt: -20
destroyed: -20
scanning: -2
good_should_be_repairing: 0.0002
good_should_be_restoring: 0.0002
good_should_be_corrupt: 0.0005
good_should_be_destroyed: 0.001
repairing_should_be_good: -0.0005
repairing_should_be_restoring: 0.0002
repairing_should_be_corrupt: 0.0002
repairing_should_be_destroyed: 0.0000
repairing: -0.0003
restoring_should_be_good: -0.001
restoring_should_be_repairing: -0.0002
restoring_should_be_corrupt: 0.0001
restoring_should_be_destroyed: 0.0002
restoring: -0.0006
corrupt_should_be_good: -0.001
corrupt_should_be_repairing: -0.001
corrupt_should_be_restoring: -0.001
corrupt_should_be_destroyed: 0.0002
corrupt: -0.001
destroyed_should_be_good: -0.002
destroyed_should_be_repairing: -0.002
destroyed_should_be_restoring: -0.002
destroyed_should_be_corrupt: -0.002
destroyed: -0.002
scanning: -0.0002
# IER status
red_ier_running: -5
green_ier_blocked: -10
red_ier_running: -0.0005
green_ier_blocked: -0.001
# Patching / Reset durations
os_patching_duration: 5 # The time taken to patch the OS

101
src/primaite/config/_package_data/training/training_config_random_red_agent.yaml
View File

@@ -1,101 +0,0 @@
# Main Config File
# Generic config values
# Choose one of these (dependent on Agent being trained)
# "STABLE_BASELINES3_PPO"
# "STABLE_BASELINES3_A2C"
# "GENERIC"
agent_identifier: STABLE_BASELINES3_A2C
# Sets whether Red Agent POL and IER is randomised.
# Options are:
# True
# False
random_red_agent: True
# Sets How the Action Space is defined:
# "NODE"
# "ACL"
# "ANY" node and acl actions
action_type: NODE
# Number of episodes to run per session
num_episodes: 10
# Number of time_steps per episode
num_steps: 256
# Time delay between steps (for generic agents)
time_delay: 10
# Type of session to be run (TRAINING or EVALUATION)
session_type: TRAINING
# Determine whether to load an agent from file
load_agent: False
# File path and file name of agent if you're loading one in
agent_load_file: C:\[Path]\[agent_saved_filename.zip]
# Environment config values
# The high value for the observation space
observation_space_high_value: 1000000000
# Reward values
# Generic
all_ok: 0
# Node Hardware State
off_should_be_on: -10
off_should_be_resetting: -5
on_should_be_off: -2
on_should_be_resetting: -5
resetting_should_be_on: -5
resetting_should_be_off: -2
resetting: -3
# Node Software or Service State
good_should_be_patching: 2
good_should_be_compromised: 5
good_should_be_overwhelmed: 5
patching_should_be_good: -5
patching_should_be_compromised: 2
patching_should_be_overwhelmed: 2
patching: -3
compromised_should_be_good: -20
compromised_should_be_patching: -20
compromised_should_be_overwhelmed: -20
compromised: -20
overwhelmed_should_be_good: -20
overwhelmed_should_be_patching: -20
overwhelmed_should_be_compromised: -20
overwhelmed: -20
# Node File System State
good_should_be_repairing: 2
good_should_be_restoring: 2
good_should_be_corrupt: 5
good_should_be_destroyed: 10
repairing_should_be_good: -5
repairing_should_be_restoring: 2
repairing_should_be_corrupt: 2
repairing_should_be_destroyed: 0
repairing: -3
restoring_should_be_good: -10
restoring_should_be_repairing: -2
restoring_should_be_corrupt: 1
restoring_should_be_destroyed: 2
restoring: -6
corrupt_should_be_good: -10
corrupt_should_be_repairing: -10
corrupt_should_be_restoring: -10
corrupt_should_be_destroyed: 2
corrupt: -10
destroyed_should_be_good: -20
destroyed_should_be_repairing: -20
destroyed_should_be_restoring: -20
destroyed_should_be_corrupt: -20
destroyed: -20
scanning: -2
# IER status
red_ier_running: -5
green_ier_blocked: -10
# Patching / Reset durations
os_patching_duration: 5 # The time taken to patch the OS
node_reset_duration: 5 # The time taken to reset a node (hardware)
service_patching_duration: 5 # The time taken to patch a service
file_system_repairing_limit: 5 # The time take to repair the file system
file_system_restoring_limit: 5 # The time take to restore the file system
file_system_scanning_limit: 5 # The time taken to scan the file system

106
src/primaite/config/training_config.py
View File

@@ -94,64 +94,64 @@ class TrainingConfig:
# Reward values
# Generic
all_ok: int = 0
all_ok: float = 0
# Node Hardware State
off_should_be_on: int = -10
off_should_be_resetting: int = -5
on_should_be_off: int = -2
on_should_be_resetting: int = -5
resetting_should_be_on: int = -5
resetting_should_be_off: int = -2
resetting: int = -3
off_should_be_on: float = -0.001
off_should_be_resetting: float = -0.0005
on_should_be_off: float = -0.0002
on_should_be_resetting: float = -0.0005
resetting_should_be_on: float = -0.0005
resetting_should_be_off: float = -0.0002
resetting: float = -0.0003
# Node Software or Service State
good_should_be_patching: int = 2
good_should_be_compromised: int = 5
good_should_be_overwhelmed: int = 5
patching_should_be_good: int = -5
patching_should_be_compromised: int = 2
patching_should_be_overwhelmed: int = 2
patching: int = -3
compromised_should_be_good: int = -20
compromised_should_be_patching: int = -20
compromised_should_be_overwhelmed: int = -20
compromised: int = -20
overwhelmed_should_be_good: int = -20
overwhelmed_should_be_patching: int = -20
overwhelmed_should_be_compromised: int = -20
overwhelmed: int = -20
good_should_be_patching: float = 0.0002
good_should_be_compromised: float = 0.0005
good_should_be_overwhelmed: float = 0.0005
patching_should_be_good: float = -0.0005
patching_should_be_compromised: float = 0.0002
patching_should_be_overwhelmed: float = 0.0002
patching: float = -0.0003
compromised_should_be_good: float = -0.002
compromised_should_be_patching: float = -0.002
compromised_should_be_overwhelmed: float = -0.002
compromised: float = -0.002
overwhelmed_should_be_good: float = -0.002
overwhelmed_should_be_patching: float = -0.002
overwhelmed_should_be_compromised: float = -0.002
overwhelmed: float = -0.002
# Node File System State
good_should_be_repairing: int = 2
good_should_be_restoring: int = 2
good_should_be_corrupt: int = 5
good_should_be_destroyed: int = 10
repairing_should_be_good: int = -5
repairing_should_be_restoring: int = 2
repairing_should_be_corrupt: int = 2
repairing_should_be_destroyed: int = 0
repairing: int = -3
restoring_should_be_good: int = -10
restoring_should_be_repairing: int = -2
restoring_should_be_corrupt: int = 1
restoring_should_be_destroyed: int = 2
restoring: int = -6
corrupt_should_be_good: int = -10
corrupt_should_be_repairing: int = -10
corrupt_should_be_restoring: int = -10
corrupt_should_be_destroyed: int = 2
corrupt: int = -10
destroyed_should_be_good: int = -20
destroyed_should_be_repairing: int = -20
destroyed_should_be_restoring: int = -20
destroyed_should_be_corrupt: int = -20
destroyed: int = -20
scanning: int = -2
good_should_be_repairing: float = 0.0002
good_should_be_restoring: float = 0.0002
good_should_be_corrupt: float = 0.0005
good_should_be_destroyed: float = 0.001
repairing_should_be_good: float = -0.0005
repairing_should_be_restoring: float = 0.0002
repairing_should_be_corrupt: float = 0.0002
repairing_should_be_destroyed: float = 0.0000
repairing: float = -0.0003
restoring_should_be_good: float = -0.001
restoring_should_be_repairing: float = -0.0002
restoring_should_be_corrupt: float = 0.0001
restoring_should_be_destroyed: float = 0.0002
restoring: float = -0.0006
corrupt_should_be_good: float = -0.001
corrupt_should_be_repairing: float = -0.001
corrupt_should_be_restoring: float = -0.001
corrupt_should_be_destroyed: float = 0.0002
corrupt: float = -0.001
destroyed_should_be_good: float = -0.002
destroyed_should_be_repairing: float = -0.002
destroyed_should_be_restoring: float = -0.002
destroyed_should_be_corrupt: float = -0.002
destroyed: float = -0.002
scanning: float = -0.0002
# IER status
red_ier_running: int = -5
green_ier_blocked: int = -10
red_ier_running: float = -0.0005
green_ier_blocked: float = -0.001
# Patching / Reset durations
os_patching_duration: int = 5
@@ -178,6 +178,12 @@ class TrainingConfig:
file_system_scanning_limit: int = 5
"The time taken to scan the file system"
deterministic: bool = False
"If true, the training will be deterministic"
seed: Optional[int] = None
"The random number generator seed to be used while training the agent"
@classmethod
def from_dict(cls, config_dict: Dict[str, Union[str, int, bool]]) -> TrainingConfig:
"""

8
src/primaite/environment/primaite_env.py
View File

@@ -141,10 +141,10 @@ class Primaite(Env):
self.step_info = {}
# Total reward
self.total_reward = 0
self.total_reward: float = 0
# Average reward
self.average_reward = 0
self.average_reward: float = 0
# Episode count
self.episode_count = 0
@@ -282,9 +282,9 @@ class Primaite(Env):
self._create_random_red_agent()
# Reset counters and totals
self.total_reward = 0
self.total_reward = 0.0
self.step_count = 0
self.average_reward = 0
self.average_reward = 0.0
# Update observations space and return
self.update_environent_obs()

20
src/primaite/environment/reward.py
View File

@@ -20,7 +20,7 @@ def calculate_reward_function(
red_iers,
step_count,
config_values,
):
) -> float:
"""
Compares the states of the initial and final nodes/links to get a reward.
@@ -33,7 +33,7 @@ def calculate_reward_function(
step_count: current step
config_values: Config values
"""
reward_value = 0
reward_value: float = 0.0
# For each node, compare hardware state, SoftwareState, service states
for node_key, final_node in final_nodes.items():
@@ -94,7 +94,7 @@ def calculate_reward_function(
return reward_value
def score_node_operating_state(final_node, initial_node, reference_node, config_values):
def score_node_operating_state(final_node, initial_node, reference_node, config_values) -> float:
"""
Calculates score relating to the hardware state of a node.
@@ -104,7 +104,7 @@ def score_node_operating_state(final_node, initial_node, reference_node, config_
reference_node: The node if there had been no red or blue effect
config_values: Config values
"""
score = 0
score: float = 0.0
final_node_operating_state = final_node.hardware_state
reference_node_operating_state = reference_node.hardware_state
@@ -143,7 +143,7 @@ def score_node_operating_state(final_node, initial_node, reference_node, config_
return score
def score_node_os_state(final_node, initial_node, reference_node, config_values):
def score_node_os_state(final_node, initial_node, reference_node, config_values) -> float:
"""
Calculates score relating to the Software State of a node.
@@ -153,7 +153,7 @@ def score_node_os_state(final_node, initial_node, reference_node, config_values)
reference_node: The node if there had been no red or blue effect
config_values: Config values
"""
score = 0
score: float = 0.0
final_node_os_state = final_node.software_state
reference_node_os_state = reference_node.software_state
@@ -194,7 +194,7 @@ def score_node_os_state(final_node, initial_node, reference_node, config_values)
return score
def score_node_service_state(final_node, initial_node, reference_node, config_values):
def score_node_service_state(final_node, initial_node, reference_node, config_values) -> float:
"""
Calculates score relating to the service state(s) of a node.
@@ -204,7 +204,7 @@ def score_node_service_state(final_node, initial_node, reference_node, config_va
reference_node: The node if there had been no red or blue effect
config_values: Config values
"""
score = 0
score: float = 0.0
final_node_services: Dict[str, Service] = final_node.services
reference_node_services: Dict[str, Service] = reference_node.services
@@ -266,7 +266,7 @@ def score_node_service_state(final_node, initial_node, reference_node, config_va
return score
def score_node_file_system(final_node, initial_node, reference_node, config_values):
def score_node_file_system(final_node, initial_node, reference_node, config_values) -> float:
"""
Calculates score relating to the file system state of a node.
@@ -275,7 +275,7 @@ def score_node_file_system(final_node, initial_node, reference_node, config_valu
initial_node: The node before red and blue agents take effect
reference_node: The node if there had been no red or blue effect
"""
score = 0
score: float = 0.0
final_node_file_system_state = final_node.file_system_state_actual
reference_node_file_system_state = reference_node.file_system_state_actual

2
src/primaite/transactions/transaction.py
View File

@@ -31,7 +31,7 @@ class Transaction(object):
"The observation space before any actions are taken"
self.obs_space_post = None
"The observation space after any actions are taken"
self.reward = None
self.reward: float = None
"The reward value"
self.action_space = None
"The action space invoked by the agent"