Code

Algorithmic Expansion to PPO#462

Code

Draft

kim-mskw wants to merge 33 commits intomainassume-framework/assume:mainfrom drl-ppoassume-framework/assume:drl-ppoCopy head branch name to clipboard

+5,356-482

Commits

Commits on Sep 11, 2024

• PPO

  

  ufqjh

  committed
  e996737

• Merge branch 'main' into drl-ppo

  

  ufqjh

  committed
  5e12c2f

Commits on Sep 20, 2024

• DRL PPO Update

  

  ufqjh

  committed
  46f843a

• Merge branch 'main' into drl-ppo

  

  ufqjh

  committed
  4b889aa

Commits on Sep 28, 2024

• Buffer changes, update_policy and further advancements in DRL related files

  

  ufqjh

  committed
  552f549

Commits on Oct 1, 2024

• first RUNABLE BUT NOT VALIDATED Version of PPO

  Show description for 128727f
  kim-mskw

  committed
  128727f

Commits on Oct 19, 2024

• - Removed unused parts of the RolloutBuffer

  Show description for 991c311
  

  ufqjh

  committed
  991c311

Commits on Oct 23, 2024

• Implemented centralized critic

  

  ufqjh

  committed
  aba37f1

Commits on Oct 25, 2024

• Fixed comments regarding centralized critic

  

  ufqjh

  committed
  9047578

Commits on Oct 28, 2024

• Merge branch 'main' into drl-ppo

  nick-harder

  authored
  a5e0f5f

• - implemnted perform eval differentiation that gets rid of stochasticity

  Show description for ca1f9fc
  kim-mskw

  committed
  ca1f9fc

• Merge remote-tracking branch 'origin/main' into drl-ppo

  maurerle

  committed
  d102a2d

Commits on Oct 29, 2024

• Merge branch 'drl-ppo' of https://github.com/assume-framework/assume into drl-ppo

  kim-mskw

  committed
  7e19389

• - tensor handling in get_actions function

  Show description for e011baa
  kim-mskw

  committed
  e011baa

• convergence of PPO? zurückhaltendes wuhu

  kim-mskw

  committed
  5d2dd2d

• - pushed config setting for reproducability

  kim-mskw

  committed
  79c287a

• - added further todos for prettier code and better handling of critic information

  kim-mskw

  committed
  97c3bb6

Commits on Oct 30, 2024

• Merge branch 'main' into drl-ppo

  nick-harder

  authored
  9d669e0

• - moved advantage and value calculation outside of the loop since it would always be calculated new with the gradient steps, which is not in accordance with the spinning up implementation

  Show description for 77f28d5
  kim-mskw

  committed
  77f28d5

• - introduce new actor architecture with distributuon layer so that we do not define two distribtuions one in get_actions and one in policiy update, as this is prone to mistakes

  kim-mskw

  committed
  a0c2a88

• Merge branch 'drl-ppo' of https://github.com/assume-framework/assume into drl-ppo

  kim-mskw

  committed
  8542f08

• - delted epochs from config

  kim-mskw

  committed
  cd30915

Commits on Nov 14, 2024

• add mini batch sampling to ppo

  adiwied

  committed
  5ad6e15

Commits on Nov 15, 2024

• fix clamping of action distribution

  adiwied

  committed
  8928bf7

Commits on Nov 18, 2024

• ppo is now stable in ex2a base, added orthogonal initialization and working hyper params

  adiwied

  committed
  949ba73

Commits on Nov 19, 2024

• improve hyperparams

  adiwied

  committed
  68c16a9

Commits on Nov 20, 2024

• Merge pull request #487 from adiwied/drl-ppo

  Show description for 8b3196a
  kim-mskw

  authored
  8b3196a

Commits on Nov 21, 2024

• Merge branch 'main' of https://github.com/assume-framework/assume into drl-ppo

  kim-mskw

  committed
  2cf7d0b

• align all configs with additional algorithm feature

  kim-mskw

  committed
  9ce1c36

Commits on Nov 25, 2024

• - adjusted all learning configs to match new config format

  kim-mskw

  committed
  3f1ac6c

Commits on Dec 3, 2024

• - cleaned weirdly merged configs

  Show description for 61e9bfa
  kim-mskw

  committed
  61e9bfa

Commits on Dec 4, 2024

• Merge branch 'main' of https://github.com/assume-framework/assume into drl-ppo

  kim-mskw

  committed
  9d43562

• - adjusted exploration noise handling to fit to PPO, makes merge commit runable

  kim-mskw

  committed
  3fa7359