Skip to content

Latest commit

 

History

History
90 lines (73 loc) · 7.55 KB

CHANGELOG.md

File metadata and controls

90 lines (73 loc) · 7.55 KB
Raw

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

Unreleased

2.0.0 - 2023-07-26

Added

  • Command Line Interface (CLI) for easy access and streamlined usage of PrimAITE.
  • Application Directories to enable PrimAITE as a Python package with predefined directories for storage.
  • Support for Ray Rllib, allowing training of PPO and A2C agents using Stable Baselines3 and Ray RLlib.
  • Random Red Agent to train the blue agent against, with options for randomised Red Agent POL and IER.
  • Repeatability of sessions through seed settings, and deterministic or stochastic evaluation options.
  • Session loading to revisit previously run sessions for SB3 Agents.
  • Agent Session Classes (AgentSessionABC and HardCodedAgentSessionABC) to standardise agent training with a common interface.
  • Standardised Session Output in a structured format in the user's app sessions directory, providing four types of outputs:
    1. Session Metadata
    2. Results
    3. Diagrams
    4. Saved agents (training checkpoints and a final trained agent).
  • Configurable Observation Space managed by the ObservationHandler class for a more flexible observation space setup.
  • Benchmarking of PrimAITE performance, showcasing session and step durations for reference.
  • Documentation overhaul, including automatic API and test documentation with recursive Sphinx auto-summary, using the Furo theme for responsive light/dark theme, and enhanced navigation with sphinx-code-tabs and sphinx-copybutton.

Changed

  • Action Space updated to discrete spaces, introducing a new ANY action space option for combined NODE and ACL actions.
  • Improved Node attribute naming convention for consistency, now adhering to Pascal Case.
  • Package Structure has been refactored for better build, distribution, and installation, with all source code now in the src/ directory, and the PRIMAITE Python package renamed to primaite to adhere to PEP-8 Package & Module Names.
  • Docs and Tests now sit outside the src/ directory.
  • Non-python files (example config files, Jupyter notebooks, etc.) now sit inside a */_package_data/ directory in their respective sub-packages.
  • All dependencies are now defined in the pyproject.toml file.
  • Introduced individual configuration for the number of episodes and time steps for training and evaluation sessions, with separate config values for each.
  • Decoupled the lay down config file from the training config, allowing more flexibility in configuration management.
  • Updated Transactions to only report pre-action observation, improving the CSV header and providing more human-readable descriptions for columns relating to observations.
  • Changes to AccessControlList, where the acl dictionary is now a list to accommodate changes to ACL action space and positioning of ACLRules inside the list to signal their level of priority.

Fixed

  • Various bug fixes, including Green IERs separation, correct clearing of links in the reference environment, and proper reward calculation.
  • Logic to check if a node is OFF before executing actions on the node by the blue agent, preventing erroneous state changes.
  • Improved functionality of Resetting a Node, adding "SHUTTING DOWN" and "BOOTING" operating states for more reliable reset commands.
  • Corrected the order of actions in the Primaite env to ensure the blue agent uses the current state for decision-making.

[1.1.1] - 2023-06-27

Bug Fixes

  • Fixed bug whereby 'reference' environment links reach bandwidth capacity and are never cleared due to green & red IERs being applied to them. This bug had a knock-on effect that meant IERs were being blocked based on the full capacity of links on the reference environment which was not correct; they should only be based on the link capacity of the 'live' environment. This fix has been addressed by:
    • Implementing a reference copy of all green IERs (self.green_iers_reference).
    • Clearing the traffic on reference IERs at the same time as the live IERs.
    • Passing the green_iers_reference to the apply_iers function at the reference stage.
    • Passing the green_iers_reference as an additional argument to calculate_reward_function.
    • Updating the green IERs section of the calculate_reward_function to now take into account both the green reference IERs and live IERs. The green_ier_blocked reward is only applied if the IER is blocked in the live environment but is running in the reference environment.
    • Re-ordering the actions taken as part of the step function to ensure the blue action happens first before other changes.
    • Removing the unnecessary "Reapply PoL and IERs" action from the step function.
    • Moving the deep-copy of nodes and links to below the "Implement blue action" stage of the step function.

[1.1.0] - 2023-03-13

Added

  • The user can now initiate either a TRAINING session or an EVALUATION (test) session with the Stable Baselines 3 (SB3) agents via the config_main.yaml file. During evaluation/testing, the agent policy will be fixed (no longer learning) and subjected to the SB3 evaluate_policy() function.
  • The user can choose whether a saved agent is loaded into the session (with reference to a URL) via the config_main.yaml file. They specify a Boolean true/false indicating whether a saved agent should be loaded, and specify the URL and file name.
  • Active and Service nodes now possess a new "File System State" attribute. This attribute is permitted to have the states GOOD, CORRUPT, DESTROYED, REPAIRING, and RESTORING. This new feature affects the following components:
    • Blue agent observation space;
    • Blue agent action space;
    • Reward function;
    • Node pattern-of-life.
  • The Red Agent node pattern-of-life has been enhanced so that node PoL is triggered by an 'initiator'. The initiator is either DIRECT (state change is applied to the node without any conditions), IER (state change is applied to the node based on IER entry condition), or SERVICE (state change is applied to the node based on a service state condition on the same node or a different node within the network).
  • New default config named "config_5_DATA_MANIPULATION.yaml" and associated Training Use Case Profile.
  • NodeStateInstruction has been split into NodeStateInstructionGreen and NodeStateInstructionRed to reflect the changes within the red agent pattern-of-life capability.
  • The reward function has been enhanced so that node attribute states of resetting, patching, repairing, and restarting contribute to the overall reward value.
  • The User Guide has been updated to reflect all the above changes.

Changed

  • "config_1_DDOS_BASIC.yaml" modified to make it more simplistic to aid evaluation testing.
  • "config_2_DDOS_BASIC.yaml" updated to reflect the addition of the File System State and the Red Agent node pattern-of-life enhancement.
  • "config_3_DOS_VERY_BASIC.yaml" updated to reflect the addition of the File System State and the Red Agent node pattern-of-life enhancement.
  • "config_UNIT_TEST.yaml" is a copy of the new "config_5_DATA_MANIPULATION.yaml" file.
  • Updates to Transactions.

Fixed

  • Fixed "config_2_DDOS_BASIC.yaml" by adding another ACL rule to allow traffic to flow from Node 9 to Node 3. Previously, there was no rule, so one of the green IERs could not flow by default.