Code

Flappy bird RL challenge#53

Code

Open

MedericFourmy wants to merge 25 commits intoSupaeroDataScience:masterSupaeroDataScience/RLchallenge:masterfrom MedericFourmy:masterMedericFourmy/RLchallenge:masterCopy head branch name to clipboard

+780

Commits

Commits on Jan 24, 2018

• Deterministic policies

  MedericFourmy

  committed
  a366a88

  Copy full SHA for a366a88

Commits on Jan 30, 2018

• Initial training algorithm: Simple Q-Learning model, does not converge in 1000 games

  MedericFourmy

  committed
  cfd2711

  Copy full SHA for cfd2711

Commits on Feb 5, 2018

• Neural Q learning implemented

  MedericFourmy

  committed
  585aee5

  Copy full SHA for 585aee5

Commits on Feb 6, 2018

• Some success with SARSA algorithm but needs to be tuned

  MedericFourmy

  committed
  d79c6d0

  Copy full SHA for d79c6d0

Commits on Feb 7, 2018

• Plain SARSA more or less working

  MedericFourmy

  committed
  24fb8bf

  Copy full SHA for 24fb8bf

Commits on Feb 9, 2018

• Average of 21 using LAMBDA SARSA

  MedericFourmy

  committed
  594af91

  Copy full SHA for 594af91

Commits on Feb 15, 2018

• Faster Eps diminution + timing: 12.5

  MedericFourmy

  committed
  38def90

  Copy full SHA for 38def90

Commits on Feb 16, 2018

• 2000000 iteration + shorter time constant for eps decreasing: mean of 45!

  Show description for 4812702
  MedericFourmy

  committed
  4812702

  Copy full SHA for 4812702

Commits on Feb 20, 2018

• After different things tried unsuccessfully, achieved a 45 mean with 2000000 iterations. Modified train to retrain based on previous training

  MedericFourmy

  committed
  63bdc35

  Copy full SHA for 63bdc35

• Lambda Sarsa --> Avg nb of doors: 70, in avg less than 1 state not seen per game. Might still be optimised but not necesseray

  MedericFourmy

  committed
  fd1414d

  Copy full SHA for fd1414d

Commits on Mar 8, 2018

• Trying scaler on Q-learning NN

  

  mfourmy

  committed
  2b62274

  Copy full SHA for 2b62274

• Added a DISPLAY PARAM

  

  mfourmy

  committed
  3e7e408

  Copy full SHA for 3e7e408

Commits on Mar 10, 2018

• No window + no scaler

  

  MedericFourmy

  committed
  4a92f62

  Copy full SHA for 4a92f62

• solved nan weights pbe

  

  mfourmy

  committed
  638f55e

  Copy full SHA for 638f55e

Commits on Mar 11, 2018

• Play and test as same function

  

  mfourmy

  committed
  ea84312

  Copy full SHA for ea84312

• More refactoring

  

  mfourmy

  committed
  bac55cb

  Copy full SHA for bac55cb

• different files

  

  mfourmy

  committed
  6ca19d1

  Copy full SHA for 6ca19d1

• Files renamed

  

  mfourmy

  committed
  a5b9146

  Copy full SHA for a5b9146

• Merge pull request #1 from SupaeroDataScience/master

  Show description for aae26e6
  

  MedericFourmy

  authored
  aae26e6

  Copy full SHA for aae26e6

• Beginning of DQL

  

  mfourmy

  committed
  6c3aa02

  Copy full SHA for 6c3aa02

• Merge branch 'master' of https://github.com/MedericFourmy/RLchallenge

  

  mfourmy

  committed
  01dad4e

  Copy full SHA for 01dad4e

• DQL implemented

  

  mfourmy

  committed
  6a5db7d

  Copy full SHA for 6a5db7d

• Fixed dumb save every time bug

  

  mfourmy

  committed
  f80b4c2

  Copy full SHA for f80b4c2

• Working LambdaSarsa

  

  mfourmy

  committed
  6480bdc

  Copy full SHA for 6480bdc

• update epsilon bug solved + doom pbe

  

  mfourmy

  committed
  4a9ac30

  Copy full SHA for 4a9ac30