Code

[RL-baseline] Model v4#38

Code

Open

ziritrion wants to merge 25 commits intoRL-with-baselinexeviknal/aidl-2021-wo-rl:RL-with-baselinefrom RL-baseline-v4xeviknal/aidl-2021-wo-rl:RL-baseline-v4Copy head branch name to clipboard

+142-68

Commits

Commits on Mar 11, 2021

• Add metrics and logsoftmax

  xeviknal

  committed
  2106b4b

  Copy full SHA for 2106b4b

• Updating the model

  xeviknal

  committed
  0f2f195

  Copy full SHA for 0f2f195

• Add new model to baseline

  xeviknal

  committed
  39b3ad3

  Copy full SHA for 39b3ad3

Commits on Mar 14, 2021

• The line that fixes all

  

  xeviknal

  authored and

  ziritrion

  committed
  907af7a

  Copy full SHA for 907af7a

• Add mean entropy - to reduce tensorboard runs

  

  xeviknal

  authored and

  ziritrion

  committed
  8e4ee6c

  Copy full SHA for 8e4ee6c

• Add action prob mean: mean of prob of actions taken in the episode

  xeviknal

  committed
  3970702

  Copy full SHA for 3970702

• Added simple directory check to params folder

  ziritrion

  committed
  957a3b4

  Copy full SHA for 957a3b4

• Added additional param save conditions (end of log_interval, last episode, successful training)

  ziritrion

  committed
  0105564

  Copy full SHA for 0105564

• Merge branch 'RL-baseline-new-model' of github.com:xeviknal/aidl-2021-wo-rl into RL-baseline-new-model

  Show description for b5a5184
  ziritrion

  committed
  b5a5184

  Copy full SHA for b5a5184

• Removing old runs; they don't apply to this branch

  ziritrion

  committed
  c6954ec

  Copy full SHA for c6954ec

Commits on Mar 15, 2021

• RL-baseline-NM-save-optim

  

  xeviknal

  authored and

  ziritrion

  committed
  a7c907c

  Copy full SHA for a7c907c

• Load optimizer params

  

  xeviknal

  authored and

  ziritrion

  committed
  d5b676c

  Copy full SHA for d5b676c

• 8k runs

  ziritrion

  committed
  bd7f6c0

  Copy full SHA for bd7f6c0

• Fresh start with latest checkpoint load-save changes. Also, small gitignore change

  ziritrion

  committed
  5f246c0

  Copy full SHA for 5f246c0

• bugfix

  ziritrion

  committed
  e8aa5e4

  Copy full SHA for e8aa5e4

• 10k runs

  ziritrion

  committed
  c52a4f2

  Copy full SHA for c52a4f2

Commits on Mar 16, 2021

• Almost 20k runs. Reward is starting to improve little by little

  ziritrion

  committed
  192bece

  Copy full SHA for 192bece

• Fixed runner.py for generating videos

  ziritrion

  committed
  11b46c4

  Copy full SHA for 11b46c4

• 25k runs. Slight improvement but far from desirable

  ziritrion

  committed
  befd201

  Copy full SHA for befd201

Commits on Mar 23, 2021

• Adding more linear layers for each head

  xeviknal

  committed
  610d5a8

  Copy full SHA for 610d5a8

Commits on Mar 25, 2021

• Cleaned up the code and removed params and runs in order to ease merging with other branches

  ziritrion

  committed
  e9ad814

  Copy full SHA for e9ad814

• Modified actions to include the no_action possibility, as well as finetuning

  ziritrion

  committed
  4409cf6

  Copy full SHA for 4409cf6

• Tweaked the way the params filename is generated, as well as tensorboard references for ease of use

  ziritrion

  committed
  db1bd06

  Copy full SHA for db1bd06

• In train_episode, moved some vars to GPU that weren't being moved before

  ziritrion

  committed
  e5a9707

  Copy full SHA for e5a9707

Commits on Mar 29, 2021

• Base start for Baseline model v4

  ziritrion

  committed
  c3aa2c7

  Copy full SHA for c3aa2c7